Ocular Responses to Motion Parallax Stimuli: The Role of Perceptual and Attentional Factors

P PII: S0042-6989(96)00314-8 VisionRes., Vol.37,No.12,pp. 1627-1641,1997 @1997ElsevierScienceLtd.All rightsreserved Printedin GreatBritain 0042-6989/97 $17.00+ 0.00 Ocular Responses to Motion Parallax Stimuli: The Role of Perceptual and Attentional Factors D. R. MESTRE,*~ G. S. MASSON* Received21June 1995;in revisedform 23April1996;infinalform 7 October1996 When human subjects are presented with visual displays consisting of random dots moving sideways at different velocities, they perceive transparent surfaces, moving in the same direction but located at different distances from themselves. They perceive depth from motion parallax, without any additional cues to depth, such as relative size, occlusion or binocular disparity. Simultaneously, large-field visual motion triggers compensatory eye movements which tend to offset such motion, in order to stabilize the visual image of the environment. In a series of experiments, we investigated how such reflexive eye movements are controlled by motion parallax displays, that is, in a situation where a complete stabilization of the visual image is never possible. Results show that optokinetic nystagmus, and not merely active visual pursuit of singular elements, is triggered by such displays. Prior to the detection of depth from motion parallax, eye tracking velocity is equal to the average velocity of the visual image. After detection, eye tracking velocity spontaneously matches the slowest velocity in the visual field, but can be controlled by attentional factors. Finally, for a visual stimulation containing more than three velocities, subjects are no longer able to perceptually dissociate between different surfaces in depth, and eye tracking velocity remains equal to the average velocity of the visual image. These data suggest that, in the presence of flow fields containing motion parallax, optokinetic eye movements are modulated by perceptual and attentional factors. 01997 Elsevier Science Ltd. Optic flow Motion perception Eye movements INTRODUCTION To an observer’s translation through the environment corresponds a complex, continuous, transformation of retinal images, which dependsboth on the characteristics of the observer’s trajectory and on the three-dimensional environmental structure (Gibson, 1954; Gibson et al., 1955). Such an optical flow field may be further complicated by the fact that at a specificretinal location, several optical motions might occur simultaneously, a situation called motion transparency. A fundamental problem in perception is understanding how the visual system segregates and groups motion signals in the optic flow field, in order to achieve the perception of a threedimensional layout. The ability to classify the differentmotion components in the optic flow field extends to the oculomotorsystem. Primates have reflexivevisual tracking systems that help stabilize the eyes in a moving visual world. Several studies in both human and monkeys have suggested that some of these reflexivevisual tracking mechanisms,also OKN Attention called optokinetic reflexes, are cortically mediated and able to deal with complex optic flow fields (Miles et al., 1986; Gellman et al., 1990; Busettiniet al., 1991, 1996). When the observer is presented with a flow field containing motion parallax, the visual stabilization of gaze must solve a “bottleneck problem” (Ferrera & Lisberger, 1995): althoughmultiple motion signals enter the visual system,they correspond,at any given moment, to a single eye movement velocity. Therefore, as the global motion of the image cannot be offset by any appropriate eye movement, the tracking system must be able to single out a particular motion signal to be cancelled. Such integration, segmentation and selection mechanisms might depend both on automatic and attentional,perception-basedprocesses. There are differentways to investigatehow the primate oculomotor system deals with visual motion in a threedimensional environment. First, several studies suggest that the visual motion-processingsystem involved in the controlof eye movementsintegratesdifferentdepth cues, in order to single out the motion of a selected object. For instance, several studies have demonstrated a link between stereoscopic mechanisms and the optokinetic system (Howard & Simpson, 1989), In both human and monkeys, this link affects automatic, low-level, motion detection, as demonstrated by the disparity-tuning of short-latencyocularfollowingresponses.The responseto *Centre de Recherches en Neuroscience Cognitive, Centre National de la Recherche Scientifique, UPR 9012, 31 Chemin Joseph Aiguier, 13402Marseille Cedex 20, France. ~Towhom all correspondenceshouldbe addressed[Tel(33) (0) 49116 43 30; Fax(33) (0) 4917749 69; Entailmestre@lnf.cnrs-mrs.fr]. 1627 1628 D. R. MESTRE and G. S. MASSON a visual scene moving outsidethe binocularlyfused plane of fixationis decreased relative to that observedwhen the movement is presented in the plane of fixation(Busettini et al., 1996). Such properties may be explained by the presence of disparity-sensitive motion detectors at various stages of the primate visual cortex (Poggio & Talbot, 1981). The optokinetic system might also use other visuo-spatial cues, such as motion boundaries or dynamic occlusion, in order to select the motion of a given depth-plane (Gellman et al., 1990). Another approach is to try to understand whether high-level, perceptual or attentional processes are involved in the selection mechanism. For instance, Kowler et al. (1984) have demonstrated that subjects are able to voluntarily pursue a moving textured field in the presence of a superimposed stationary field. However, most of these studies have been concerned with the ability to select the local motion signal driving voluntary smooth pursuit eye movements and to ignore other motion signals arising from the moving visual surroundings during pursuit (Collewijn & Tamminga, 1984;Yee et al., 1983;Masson et al., 1995). In particular, these studies do not provide information about how reflexive eye movements are controlled in complex flow fields. The specificaim of the present work was to investigate whether the control of reflexivetracking eye movements depends on the visual processing of differential motion parallax and/or on the perception of three-dimensional structure from motion. Many studies have demonstrated psychophysically that the segregation of different surfaces lying at different distances from the observer can be based on the perception of velocity differences (Baker & Braddick, 1982; Braunstein, 1966; Rogers & Graham, 1979;Braunstein& Andersen, 1981;Braunstein & Title, 1988; Andersen, 1989). Most of these studies used random dot optical flows, in which no other cues to depth than differential motion parallax were present. A classical method to build such displays is to randomly position dots throughout the visual field and to set them into motion at different velocities, randomly distributed across the dots. Such displays generate the perception of moving elements lying at different distances from the observer. They also lead to the perception of motion transparency. For instance, when a subject faces a stimuluswhere two motion signals occur at each specific retinal location, (s)he usually reports the perception of two independent transparent surfaces sliding over each other. Moreover, the fact that subjectsusually report that one surface (the one which is associated with faster moving dots) is seen in front of the other, suggests a functional link between the perception of transparency and the perception of depth from motion (Kersten et al., 1992). In the following experiments, we investigated both psychophysicaland oculomotorresponsesto transparent, motion parallax stimuli. Optical flows were computerdesignedand projected on a tangent screen, in such a way that binocular disparity or distance informationremained constant across conditions. Such displays always trig- gered optokinetic nystagmus (OKN), consisting of a successionof tracking eye movementsin the direction of visual motion (slow-phase of OKN) and fast resetting saccades in the opposite direction. We used the characteristic slow temporal build-up of the perception of structure-from-motion(Treue et al., 1991) to explore the propertiesof tracking eye movementsduring the first, initial eye velocity rise, that is before the perception of structure-from-motionoccurred, and during the steadystate OKN triggered by long-lasting stimuli, when subjects were asked to pay attention to the global optic flow or to a given component in the velocity flow field. Tracking eye movementswere analysed during the slow phase of OKN. GENERALMETHODS Visual stimuli Random dot optical flows were computer generated using a micro-computer (HP 486 DX2, 66 MHz) expanded with an image processing system (Matrox Board SM 1281, 1280x 1024 pixel resolution). The stimuli were projected on a tangent, flat white screen using a trichrome video projector (Electrohome 3001) at 60 Hz frame rate. The flat screen was homogeneously white without any visible textural structure.The visible display size was 3.2 m horizontally and 2.6 m vertically, and straight ahead of the subjects’eyes. Subjectswere seated in front of the screen, at a viewing distanceof 3 m. At this distance, the visual stimulation covered a rectangular visual field sustaining 56 x 47 deg along the horizontal and vertical meridians, respectively. Subjects were placed in total darkness, inside a large black booth. Except for velocity field structure, all physical parameters (contrast, luminance,...) remained identical duringthe experiments.Opticalflowpatternsconsistedof rightward motion of a set of 100 randomly distributed dots. Except for the experiment using limited dot lifetime, as soon as a dot disappeared on the right side of the screen, it was replaced on the left end side. Each dot covered 0.044 deg of visual angle. Dot density was 0.04 dots/deg2.Luminance of the dots was 4 cd/m2 and background luminance was 0.002 cd/m2. Four different types of optical flow patterns were used across the different experiments. All consisted of a translational flow field where the number of motion signals in the velocity field, the magnitudeof velocity differences and, in one experiment,dot lifetimewere manipulated.Vision was always binocular. Neither texture gradients (change in dot size nor dot density with simulated distance from the.observer),binoculardisparity(all the dots were in the same actual plane of the screen) nor dot luminance were manipulated.Single-velocityoptical flows consisted of a homogeneous rightward motion of all the dots, at constant linear velocity. Single-velocity flow fields resulted in the perceptionof a rigid vertical plane moving right:wardat constant velocity. The different velocities used in single-velocityflows were combined to generate either double-velocity or triple-velocity flows. Double- EYE MOVEMENTSAND MOTION PARALLAX 1629 c B A -r I mdeg -L e 1 e L I ,’ 20 deghec 5 J=’=’’====~ ~ I I I I I 1 [ o 10 20 TIME [see] 30 40 o 10 t 20 TIME [WC] I I 1 1 I I 30 40 0 10 20 30 I 40 TIME [see] FIGURE 1. Eye movements elicited by (A) an optical flow with a single visual velocity (21 degkec) specifying rightward motion of a vertical plane; (B) an optical flow containing three velocities (5, 11 and 21 deg/see), specifying three moving vertical surfaces, separated in depth. The subject was asked to pay attentionto global motion.(C) The same triple-velocityflow when the subject was asked to pay attention to the surface specified by a velocity of 21 deg,kec,as indicated by the arrow. e, horizontal eye position; g horizontal eye velocity; s stimulus velocity. Velocity of saccadic eye movements is clipped. (triple-)velocity flowfieldsresulted in the perceptionof a rigid structure composed of two (three) vertical planes lying at two (three) distances from the observer, with distance being inversely related to velocity. Finally, multi-velocity flows were generated by having dot velocity ranging from a lower to an upper boundary. When facing such optical stimulation,subjectsperceived a moving cloud of dots extending in depth from them. The different velocities were equally and randomly attributed to the dots in the display. the case, captors were readjusted and the calibration procedure was repeated until satisfaction of the linearity criterion. In one experiment, subjects were asked to identify the 3D structurespecifiedby the visual stimulation.Reaction time was measured by recording the response of the subject on a three button computer-mouse, at 60 Hz (locked to frame rate). Data recording Eye movements were recorded with an infrared reflectionsystem (Iris Skalar Biomed. Inc.). Eye position signals were low-pass filtered (DC–1OOHz, –3 dB), digitized using a 12-bit A–D converter, collected at a frequency of 250 Hz and stored for off-line analysis. Although vision was binocular, only horizontalpositions from the left eye were recorded. Head stability was achieved using a bite bar and an impression of the subject’steeth made with a dental impressioncompound. Before each experimental session, the eye movement recording system was calibrated by having the subject look at ten different targets located at known positions (from –25 to 25 deg, with 5 deg steps), presented in random order. Linearity of the conversion from inputs from the A–D converter to angular values was checked before the beginning of each session (the correlation coefficient had to be greater than 0.995). If this was not Method Visualstimuli. Two conditionsof optic flow were used in this experiment. Single-velocity flows were used as control stimulito investigatethe gain and the distribution of slow-phaseeye movementsevoked by a stimuluswith a single motion vector. There were five conditions of single-velocity flow: 5, 11, 21, 42 and 84 deglsec (“straight ahead” optical velocities). Three conditions of triple-velocityflowswere used. Velocitieswere either 84-42-21deg/see,42-21-11 deg/secand 21-11-5 deg/see, respectively. Each of these eight conditions was presented twice. Single- and triple-velocityflow conditions were randomly interleaved. In each trial, a stimulus remained stationaryfor 1 sec and then moved for 30 sec. The screen was then blanked out for 5 sec during which subjects remained in total darkness. Subjects and procedure. Five young subjects, aged between 19 and 27 years, participated in the experiment. EXPERIMENT1 1630 D. R. MESTREand G. S. MASSON None had a history of necrologic or ophthalmologic disease, and they were all emmetropic according to the Snellen’svisual acuity test. Three subjectswere familiar with oculomotor experiments, but they were all naive regarding the purpose of the present study. All subjects gave their informed consent before the experiment. Instruction was given to the subjects to pay attention to the global visual motion without attempting to pursue a particular feature in the visual scene. Data analysis. Saccadic eye movements were automatically discarded from the eye position data by a computer algorithmusing an accelerationcriterion.Mean eye tracking velocity was computed by fitting a linear regression across the successive eye positions between two saccadic eye movements. The slope of this linear regression was an estimate of the mean eye velocity. Mean eye velocity and middle of the time-interval between the two saccades were computed for each slow-phase of OKN. These data were then gathered across a number of slow phases and over trials of similar stimulus conditions.A mean slow-phasevelocity and its standard deviation was computed for each subject and each condition. However, as illustrated in Fig. 1, OKN triggered by a triple-velocityflow typicallyconsistedof a succession of slow-phase eye movements with different velocities. To assess the contribution of each motion vector on the control of the steady-state OKN, we computed, for each subject and each condition, the distribution of slow-phase velocities (expressed as a percentage of the total number of slow phases in a given condition).Individual distributionswere averaged across subjects, in order to obtain a mean distributionprofilefor each experimental condition. In such distributions, the spreading of the distribution is dependent upon both inter-subject variability in the average eye velocity of slow phases and intra-subject variability across eye velocity during each given trial. Flow velocity [deg/see] 5 I 21 42 1 I 84 I ,, , , A r 1 .J.,.-, : , , , , : , : , n , , , , , n n , , n , : , >i ,,-. “, “*h I 1 100 Slow-phase o 0 velocity [deg/see] B /“ “1 I I Results Figure 1(A, B) illustratestypical optokineticresponses recorded in one subject facing a single- and triplevelocity flow, respectively.We can see that a 21 degJsec single-velocity flow elicited a strong, regular, OKN where eye tracking veIocity closely matched stimulus velocity. When the subject was presented with a flow field composedof three motion vectors (21,11 and 5 deg/ see), eye movements consisted of an irregular OKN. As illustrated by the velocity profile, slow-phase velocity was more irregular, with a clear dominance toward the slowest velocity in the motion parallax field. Optokinetic responsesto single-velocityflows. Singlevelocity stimuli elicited regular, consistent involuntary OKN with slow-phaseeye velocity close to flow velocity for velocities up to 40 deg/sec. Figure 2(A) illustrates average distributionprofiles obtained for the five optical flow velocities. For a velocity of 5 deg/see, the peak of the distributionwas centred on the optical flow velocity. For velocities of 11 and 21 deg/see, the peak of distribution was shifted towards lower velocities and 11 I I I 11[ I I I I I I 10 Ill 100 Flow velocity [deg/see] FIGURE 2. Optokinetic response to a single-velocity flow. (A) Distributions of the occurrence of slow-phase (SP) velocity (expressed as a percentage of the total number of slow phases, averaged across subjects), as a function of SP velocity (in deg,kec),for five different stimulusvelocities. (B) Average slow-phasevelocity, for each subject, as a function of stimulus velocity. The dotted line correspondsto an optimalratio of 1between eye and stimulusvelocity. the spreadingof the distributionincreased. For a velocity of 42 deg/see, the distribution peak is barely visible (around 30 deg/see) and the distributionspreads from 10 to 40 deg/sec. Finally, for a velocity of 84 deg/see, no peak was evident and the distribution spreads from 5 to 60 degJsec. An analysis of variance conducted on slow-phase average velocity revealed a significant effect of flow velocity (F’[4,16] = 13.73; P < 0.001), meaning that average slow-phase velocity increased as a function of 1631 EYE MOVEMENTSAND MOTION PARALLAX Flow velocities [degkec] Flow velocities [deg/see] , ~ B I I I Slow-phese +-m 1 I I I 1111 1 10 Slow-phase velocity [deg/see] 100 velocity [deg/see] Flow velocities [degkec] / / ‘-; / ~ :- C ~ s“ g :- = ~ .f $m z T$ g c / 0 0 / / / / / ,/ / / 0 n OY a fQ - / / 0 / /“ / 0“ 0 go :* g ; 0 D % RN < cu– / I I I I I I I II I 1 0 / / / I 10 Slow-phase / 100 velocity [deg/see] (5,11,21) (11 ,21 ,42) (21,~2,84) Flow velocities [degkec] FIGURE3. Optokineticresponseto a triple-velocityflow.Distributionsof the occurrenceof slow-phase(SP)velocity (averaged across subjects), as a function of SP velocity (in deg/see), when subjects were simply asked to pay attention to global motion, when presented with velocities of 5, 11 and 21 deghec (A); of 11,21 and 42 deg/sec (B); and of 21,42 and 84 deg/sec (C). (D) Average slow-phasevelocity, for each subject, as a function of optical flow velocities. Dotted lines correspondto a ratio of 1 between slow-phase eye velocity and each of the optical velocities in the optic flow. From the bottom of the figure, the three dotted lines correspond to an optimal ratio of 1 between eye velocity and the lowest, intermediate or highest velocity, respectively. optical flow velocity. Figure 2(B) shows that average slow-phase velocity was close to stimulus speed for velocitiesup to 21 deghec. For highervelocities,average slow-phase eye velocity was less than optical flow velocity. Notably, the difference between 42 and 84 deg/sec was no longer significant(P > 0.60). Figure 2(B) also shows that inter-subject variability increased for velocities greater than 40 deg/sec. An analysis of variance conducted on the standard deviation of slowphase velocity (calculated for each subject and each condition) also revealed a significant effect of flow velocity (F[4, 16] = 8.39; P < 0.01), meaning that intrasubject variability of slow-phase eye velocity increased when optical flow velocity increased. From these results, we can note that the evolution of distribution profiles associatedwith increasesof opticalflowvelocityis due to a weakening of the oculomotor response to optical flow for velocitie~greater than 40 deg/secand to an increase in inter- and intra-subject variability in slow-phase eye velocity. Optokinetic responses to triple-velocityflows. When subjects were simply asked to pay attention to global motion in a triple-velocityflow, slow-phase eye velocity distribution profiles were centred around the slowest velocity in the optical flow [Fig. 3(A–C)]. However, the spreadingof distributionprofileswas always significantly larger for triple-velocity flows than for single-velocity flows of comparable (slowest) velocity [compare Fig. 3(A-C) and Fig. 2(A)]. Concerningaverage slow-phasevelocities [Fig. 3(D)], an analysis of variance revealed that, with a triple-plane flow with velocities equal to 5, 11 and 21 deg/see, the 1632 D. R. MESTREandG; S. MASSON. average slow-phasevelocitywas significantlyhigherthan that observed with a single-plane flow with a velocity equal to 5 deghec (F’[1,4] = 47.32; P < 0.002), not significantly different from those observed with a single-plane flow with a velocity of 11 deg/sec (P> 0.90), and significantly lower than that observed with a single-plane flow drifting at 21 deg/sec (F[l, 4] = 151.9; P < 0.001). With a triple-plane flow includingvelocitiesof 11, 21 and 42 deg/see,the average slow-phasevelocitieswere significantlyhigherthan those observed with a single-plane flow of 11 deg/sec (F’[1,4] = 23.55; P < 0.008) and significantly less than those observed for a single-planeflow drifting at 21 degl sec (F[l, 4] = 744.88;P < 0.0001). Finally,with a tripleplane flowwith velocities equal to 21,42 and 84 de~sec, the average slow-phase velocities were not significantly higher than that observed with a single-plane flow moving at 21 deg/sec (F[l, 4] = 6.77; P > 0.05). These results indicate that, when subjectswere simply asked to pay attention to a display containing motion parallax, slow phases of OKN were not controlledby the average velocity of the motion parallax flowfield. On the contrary, results show that slow-phase tracking eye movements were controlled by motion components present in the flow field, with a strong bias toward the slowest velocity. However, residual retinal velocities appear to trigger some faster slow phases, increasing the average slow-phase velocity. This argument is also supported by the increased spreading of the distribution profiles,when comparing ocular responsesto single- and triple-velocity flows. EXPERIMENT2 A 0 — 40 20 . B /’ 83 msec ■ 166 msec A infinite // /’ // // /’ / //’~ 20 Single-veloeity 40 [degkec] — Triple-velocity FIGURE 4. Effects of dot lifetime on slow-phase duration (A) and velocity (B) of OKN (averaged across subjects) for single-velocity flows, for velocities of 12, 23 or 46 deg/sec and for a triple-velocity flow containing these three velocities. In this control experiment, we tested the nature of tracking eye movementsobserved in the first experiment. Were they driven by global motion or voluntary tracking eye movements of single elements in the display (one of velocity flow) or only one of these velocities (singlevelocity flow). Dot lifetime conditions and motion the slowest dots in a triple-velocityflow for instance)? conditions were randomly interleaved. StimuluspresentMethod ation, eye movement recording and data analysis were The experimental conditions were similar to Experi- the same as in Experiment 1. Subjects and procedure. Four new naive subjects, all ment 1, except that three differentlifetimedurationswere used. In the first condition,reproducingthe conditionsof emmetropes according to Snellen’s test, participated in Experiment 1, lifetime was “infinite”: any single dot the experiment.As in Experiment 1, they were asked to lasted for its whole displacementfrom the left side to the pay attention to global motion. If the subjects actually right side of the screen. In this case, the lifetime of each tracked global motion, no main effect of dot-lifetime on dot depended on its optical velocity. For a maximal the oculomotor behaviour was expected. A slight velocity of 46 deghec, dot lifetime was approximately decrease in the gain of the slow-phase responses might equal to 1 sec. Otherwise, each dot was displayed for a be expected due to the noise added in the display by pre-selected duration, defined as its lifetime (83 or decreasing dot lifetime. On the contrary, if the subjects 166 msec, that is 5 or 10 frames at 60 Hz). At the end of voluntarilyselected and tracked a given dot, the duration its lifetime, a dot disappeared and randomly reappeared of tracking eye movements should be related to dot at a new location on the screen to begin a new trajectory. lifetime. Dot lives were interleaved, such that, if dot lifetime was equal to x frames, l/x of the dots were replaced on each Results For infinite lifetime conditions, as in Experiment 1, frame. This method rendered highly improbable the persistence of a spatial pattern of dots over successive subjects correctly perceived the three-dimensionalstrucframes. Subjects were asked to stare at a display ture specified by the visual stimulation. However, the containing dots moving at 12, 23 and 46 degksec(triple- manipulationof dot lifetime clearly decreased the signal- 1633 EYE MOVEMENTSAND MOTION PARALLAX Flow velocities [deg/eec] Flow velocities [degkec] 11 5 21 11 21 42 B , ,, J ,,” h 1 , , ,, 1 , . . I # Slow-phase , , , I 100 10 /: -d! i 100 10 Slow–phase velocity [deg/see] veloeity [decjsec] Flow velocities [degkec] 21 1 42 I 84 I D ‘c 0 N 3 U-J 0 U-I { J 1 Slow-phaee .d 10 100 velocity [degkec] I (5,11,21) I (11,21,42) I (21,42,84) Flow velocities [degkec] FIGURE5. Distributionsof the percentage of occurrence of slow-phase(5P) velocity (averaged across subjects), as a function of 5P velocity, when observersare presentedwith triple-velocityflows,and askedto look successivelyat each plane specifiedby the motion field. (A) The optical flow field contains optical velocities of 5, 11 and 21 deg/sec. The three distributions(from left to right) correspond to conditions where subjects are asked to look at the farthest plane (5 deg/see), the intermediate plane (11 deg/see) and the closest plane (21 degkec). (B) The optical flow field containsvelocities of 11,21 and 42 degAec.(C) The optical flOW field contains velocities of 21, 42 and 84 degkec. (D) Average slow-phasevelocity, averaged across subjects, as a function of optical flow velocities. From bottom to top, subjects are asked to look at the farthest (slowest), intermediate and closest (fastest) plane, respectively. Vertical lines indicate standard deviation. to-noise ratio in the display. With a 166 msec dot lifetime, subjects still perceived the three-dimensional structure. Such perception was lost for the 83 msec dot lifetime condition. Figure 4 illustrates the effects of lifetime on the average slow-phase duration and slow-phase velocity across subjects. As shown in Fig. 4(A) and (B), decreasing dot lifetime resulted in a significantincrease in slow-phaseduration(F’[2,6] = 9.06,P < 0.02) and in a significant decrease in slow-phase tracking velocity (17[2,6] = 8.36, P < 0.02). Such an increase in slowphase duration when dot lifetime is reduced cannot correspond to voluntary tracking of individual dots, which would result in the opposite effect. Post-hoc analyses revealed that mean slow-phase velocity was only significantlyreduced for a dot lifetime of 83 msec, as compared to the two other lifetime values (F[l, 3] = 20.66, P < 0.02). As can be seen in Fig. 4(B), this effect was maximal for a triple-velocity, in which the reduction in slow-phasevelocity was equal to 40%. EXPERIMENT3 In Experiments 1 and 2, we demonstrated that, when subjects were simply asked to stare at motion parallax displays, tracking eye movementswere driven by global motion and spontaneouslymatched the slowest velocity 1634 D. R. MESTRE and G. S. MASSON component in the optic flow. However, the selection of a specificglobal motion signal might also be modulatedby attention. We already know that optokinetic eye movements depend on the instructions given to the subject before the stimulus onset (Honrubia et al., 1968). Thus, selective attentionalprocessesmay determinethe motion signal to be tracked in motion parallax displays. We investigated this point by asking the subject to pay attention to a given “depth plane”, perceived from the velocity field. The same three triple-velocityflows as in Experiment 1 were used. The procedure, data recording and analysis were also identical. “Planes of attention” were defined as the closest, the middle or the farthest plane in the display, corresponding to the fastest, the intermediate and the slowest velocity field, respectively. Each triple-velocity flow was presented six times (3 “attention” conditionsx 2 repetitions) in random order, for 30 sec. Subjectswere the same as in Experiment1. They easily detected the different depth planes. No subject reported failure to select a plane and to direct his/her attention towards it. Results As illustrated in Fig. l(C), when the subjects were instructed to pay attentionto a particular depth plane in a triple-velocityflow, steady-stateOKN was still observed, with slow-phaseeye velocity matchingthe velocity of the selected plane. Therefore, the average distribution of slow-phase eye velocity was centred on the velocity of the selected plane, for optical velocitiesup to 40 degkec, as illustrated in Fig. 5(A–C). This pattern of data was greatly similar to that observed when subjects were viewing single-velocity flows drifting at a similar velocity (compare Figs 5 and 2). Consequently,average slow-phase eye velocity of the OKN was close to the velocity of the selected plane, for velocities up to 40 deglsec [Fig. 5(D)]. Interestingly, when subjectswere instructedto look at the plane defined by the slowest velocity in a triple-velocity flow with velocities of 5, 11 and 22 deghec, the average velocity was significantly less than that observed when subjects had to stare at the same triple-velocity flow (F[l, 4] = 9.79; P < 0.04) while paying attention to global motion (see Experiment 1), but was still significantly greater than that observed when subjects looked at a single-velocity flow drifting at 5 degk.ec (F[l, 4] = 12.6;P < 0.02) [compare Figs 7(D), 4(D) and 3(B)]. For velocities greater than 11 deghec, the average slow-phase velocity observed when subjects looked at a given plane was not significantly different from that observed when subjects looked at a single-velocityflow of similar velocity. This result indicates that even if residual retinal velocities of the two other planes might trigger some faster slow phases when subjectsfocus their attention on the slowest plane, this is only true for a low velocity of 5 deg/sec and this effect is significantly reduced, as compared to the situation where subjects simply looked at a triple-velocityflow. EXPERIMENT4 In this experiment, both oculomotor behaviour and perceptual discriminationof structure-from-motionwere investigated simultaneously. We designed a simple experimental task in which subjects had to discriminate between optical flows, specifying either a single vertical plane (single-velocity flow), two vertical planes separated in depth (double-velocityflow) or a “cloud” of dots extending in depth (multi-velocity flow). Numerous studies have already demonstrated the ability of human observers to perceive depth from motion accurately in transparent dot-displays.To do so, subjects must extract the depth structure from relative global motions. However, it mightbe possibleto perform a discriminationtask using only local motion cues. The subjects may look for the presence of more than one velocity in a small local region of the display. To avoid this potential artifact, we interleaved double-velocity flows, in which only two velocitieswere presentand multi-velocityflows,in which more than 10 different velocities were present. In these two flows, there was more than one velocity in a local region. However, and also because we used low density random dot kinetograms, the correct discrimination between a cloud of dots and two planes required differential global motion analysis. Method Subjects were presented with three different types of optic flow field: single-velocity, double-velocity and multi-velocityflows.Six single-velocityflowswere used: 9, 12, 17, 23, 35 and 46 deg/sec. Three different doublevelocity flows were used: (9, 12), (17, 23) and (35, 46) deg/sec. Finally, a multi-velocityflow was generated, in which velocities ranged from 9 to 46 deglsec, with at least 10 steps of velocity, randomly distributed over the random dot pattern. Subjects were asked to discriminate between one plane, two planes or a cloud of dots by depressingone of the three buttons of a computer mouse. Visual stimuli were presented stationary for 2 sec. Motion was always to the right. Visual stimulation and data acquisitionwere stopped 3 sec after the discrimination of structure-from-motion,and the screen went blank. Inter-trial delay was approximately 5 sec. The three different types of flowswere interleaved during 5 blocks of 72 trials.To ensureequi-probabilitybetween each type of flow, each single-velocity flow was presented three times, each double-velocityflow was presented six times and the multi-velocity flow was presented 18 times, across one block. Data analysis. Reaction time for the discrimination task was definedas the time elapsingfrom the onset of the motion to the onset of the psychophysicalresponse. Eye movements were recorded monocularly for the entire trial. Two successive analyses were applied to eye movement recordings. First, the initial rise of eye velocity was analysed. Eye position data were digitally low-pass filtered and then differentiated.All the velocity traces for a single condition were displayed with a videographic interactive program. All the trials contain- EYE MOVEMENTSAND MOTION PARALLAX 1635 ● single-velocity Adouble-velocity ............................. .....................................■ multi-velocity t [3 ::::::::::::::. - ... ......................................... .......... . s ~. B A //46 L#. / // )/’ (35,46) //’35 f+ .a 8 ~ k3$! j I““”””””””’”””””””’-””““”””””””””-+””””””-” “ ++-+Ln- .. . .. .. . .. .. . ... .. . . .. .. .. . .. . .. . .. .. . . . . . .. . . .. . .. . .. .. . . . . . . e f ,46 /[ / 1 /1 10 degk,ec / /1 / e , o 100 I 200 300 1 400 TIME [msec] ~E-3” . . Fiowvelocities[deg/sse] FIGURE6. Initial oculomotortrackingbehaviorand psychophysicalreaction time inducedby single-, double-ormulti-velocity flows. (A) Initial eye velocity profiles for one subject. Each curve represents the average of 1040 trials. In the upper plot, the initial eye velocity elicited by flow fields containing two velocities lies between the eye velocities elicited by each of their velocity components.In the lower plot, the initial eye velocity elicited by a multi-velocity flow containingvelocities ranging between 9 and 46 deg/sec lies aroundthe averagevelocity of responseselicited independentlyby its velocity components.Only the responses to the highest and slowest velocity are represented. (B) Peak eye velocity reached by the eye 400 msec after the onset of stimulation (averaged across subjects, t SD), as a function of flow conditions.Horizontal dashed lines correspondto each of the peak velocities reached with single-velocity flows. They are used as reference lines for the peak eye velocities obtained with double-velocityor multi-velocity flows. (C) Average reaction time (averaged across subjects, t SD) obtained when subjects were asked to discriminatebetween one surface (specifiedby a single-velocityflow),two surfaces (specifiedby a double-velocityflow) and a cloud of dots extending in depth (specified by a multi-velocityflow). ing a saccadic intrusion between 50 msec before and 400 msec after the stimulus onset were discarded. This method ensured both no contamination of the velocity profiles by micro-saccadic eye movements and a fixed time window independentof the time of the first saccadic eye movement. Because we were interested in the time required for the visual processing of the optic flow, this time-window method was more accurate than the measurement of the velocity reached just before the occurrence of the first saccadic eye movement. Unfortunately, because of the poor dynamic resolution of the infra-red recording method, velocity analysis during the very first,open-loop,part of the oculomotorresponsewas impossible.After deletion of the selected trials, velocity profiles were averaged to obtain a mean velocity profile for each subject and each condition. Quantitative measureswere obtainedfor the maximal velocityreached during the initial eye velocity rise in the same time window (–50-400 msec), for each trial. The mean and standard deviation of the peak velocity were computed for each subject and for each condition.Changes in slowphase eye velocity during a trial were further investigated by computing the mean velocity of each slow-phase occurring between the onset of the first saccadic eye movement and the end of the stimulus, 3 sec after the psychophysicalresponse. Subjects. Three subjects, including the two authors, participated in the experiment. The third subject was unaware of the purpose of the experiment. Results Reaction time for structure-from-motiondiscrimination. The three subjects tested were able to discriminate between the three types of optic flow at the 90-100% correctlevel. As previouslyreported in numerousstudies, the discrimination of structure-from-motion requires a 1636 D. R. MESTREand G. S. MASSON long processing time as illustrated by the long reaction times [longer than 1 see, Fig. 6(C)]. Mean reaction time for the identification of a single plane ranged between 1.64 ~ 0.3 and 1.09 + 0.13 sec. Average reaction time across subjectswas 1.3 + 0.18 sec and no significant differences were found between the different velocities in single-velocity flows. Reaction times were significantlyshorter for single-velocitythan for double- or multi-velocity flows. Average reaction time across velocities was 1.62 t 0.28 sec for a doublevelocity flow, and 1.51 t 0.26 sec for a multi-velocity flow. No significant differences in reaction time were found between the different velocities in the doublevelocity condition or between the double- and multivelocity conditions. Data from a fourth, naive subject, obtained without the recording of eye movements, showed a similar pattern, with long reaction times and high correct response levels. Initial veloci~ rise of optokinetic eye movements. Figure 6(A) illustratesthe averagevelocity profilesof the initial tracking phase of OKN for one subject. Similar profiles were observed for the two other subjects. For a stimulus with more than one velocity, the initial velocity rise of OKN was intermediate between those evoked by the same velocities presented separately. For instance, 400 msec after the onset of a double-velocity stimulus with two velocities of 35 and 46 deg/sec [Fig. 6(B)], eye velocity was intermediate between the eye velocities evoked by single-velocity stimuli of 35 and 46 deg/sec. Similarly, with the multi-velocity stimulus, eye velocity after 400 msec was intermediate between the velocities evoked by the different single-velocityflowsmoving at a velocity ranging between 9 and 46 deg/sec. For each subject and each trial, the peak eye velocity between –50 and 400 msec relative to the onset of the random dot kinetograms was computed. Figure 6(B) illustrates the mean ( + SD) of the peak velocity for each condition. As illustrated, maximal velocity during the initial oculomotorresponsewhen the visual stimuluswas either a double- or multi-velocity flow was roughly the average of the maximal velocities reached by the eye when the subjects faced the different corresponding single-velocity flows. For instance, with a doublevelocity flow including velocities of 46 and 35 deglsec, the initial maximal eye velocity was significantly less than the maximal eye velocity observed when subjects were presented with a single-velocity flow drifting at 46 deg/sec (F[l, 2] = 46.25; P < 0.02) and significantly greater than the peak eye velocity evoked by a singlevelocity flow of 35 deg/sec (F[l, 2] = 29.79; P < 0.03). Similarly, for the multi-velocity flow, the maximal eye velocity was on average equal to 28 ~ 4 deghec, and not significantly different from the maximal eye velocities evoked by single-velocitystimuli of either 17 or 23 degl sec (P > 0.10). Time-course of optokinetic eye movements. After the initial phase of the ocular response, eye movements consisted of a regular OKN. After the psychophysical response was established, tracking eye velocity tended, --- ---- ---- --- — single-velocity —— double-velocity ---- multl-veloeity --- ---- ---------------------------------------------- (9.....46) .—-— — -- 0 12 (35,46) —_———————————— —————— (17,23) (9,12) 9 i 1 Time [see] FIGURE 7. Schematic representation of the evolution of slow-phase velocity as a function of time and of the different flow types. Regression lines have been fitted through the actual data. Regression coefficients are significant only for double-velocity flows, meaning that eye velocity is then reduced over time. For single- and multivelocity flows, regression coefficients are not significant(eye velocity remains more or less constant over time). for double-velocity flows, to be close to the slowest velocity in the flow field. Thus, for the first doublevelocity stimulus (35 and 46 deg/see), eye velocity was significantlyless than that observed when subjects were presented with a single-velocity flow of 46 deg/sec ($’[1,2] = 44.79;P < 0.02) and not significantlydifferent from that observed when subjects were presented with a single-velocity flow of 35 deg/sec (F[l, 2] = 8.17; P > 0.10). Similarly, with the second double-velocity stimulus (17 and 23 deghec), eye velocity was significantly less than that observedwith a single-velocityflow of 23 deg/sec (F[l, 2] = 59.51; P < 0.02) and not significantlydifferent from that observed with a 17 deg/sec single-velocity stimulus (F[l, 2] = 2.13; P > 0.20). For the third double-velocity stimulus, slow-phase eye velocity was between those observed with a 12 and a 9 deghec single-velocity flow, and not significantly different from either (P> 0.05). Finally, with the multi-velocity stimulus, average eye velocity was 19 t 4 deghec, and not significantlydifferent from that observed with either a 17 or a 23 deg/sec single-velocity flow field (P> 0.15). We further investigated the change in slow-phase velocitybetween the firstsaccadic eye movement and the end of stimulation. By pooling all individual data and fitting a linear regression to eye velocities plotted as a function of time (for a duration of about 4.5 see, between stimulus onset and 3 sec of stimulus duration after the subjects’response(occurring at about 1.5 see), we found that the only significant correlation coefficients were found with double-velocity flows (P< 0.02 or better). Regression coefficients were not significant for single- EYE MOVEMENTSAND MOTIONPARALLAX and multi-velocity flows. This pattern of results demonstrates that, with double-velocity flows, eye velocity decreased over time to finally match the slowest velocity in the flow field, whereas slow-phase eye velocity remained constant and equal to the average velocity of the stimulus with either single or multi-velocity flows (Fig. 7). 1637 A 12 degkec ——. 23 degkec ......... 46 degk+ec EXPERIMENT5 In the fourth experiment,we demonstratedthat, before the detection of structure from motion, the velocity of tracking eye movements equalled the average velocity of the flow field. After psychophysical detection occurred, eye velocity decreased to reach the slowest velocity in stimuli containing two velocities and specifying two planes in depth. This decrease was not observed with multi-velocity flows which contained more than ten velocities. One might then ask whether it takes longer for the oculomotor system to move down to the slowest velocity when many motion signals are present, or whether the fact that slow-phase eye velocity remains equal to the averagevelocity is related to the failureof the system to segregate the optic flow into its motion components. In the latter case, this result might be related to the limited abiIity of the human visual system to represent several motion signals simultaneously. Andersen (1989) demonstratedthat subjectswere unable to perceive more than three different motion-defined surfaces.To investigatewhether this constraintextendsto the ocnlomotor system, we recorded eye movements elicited by four different types of random dot kinetograms: single-, double-, triple- and multi-velocityflows. Method Visual stimuli and procedure. Random dot kinetograms were computed and displayed exactly in the same way as in the previous experiments. The four different types of optic flow were randomly interleaved and displayedfor 30 sec after 1 sec of stationarypresentation. At the end of the stimulus,the screen was blanked out for 5 sec before the beginning of the next trial. Three different single-velocityflowswere used as controls (46, 23 and 12 deghec). The triple-velocity flow was composed of these three motion components randomly distributed over the random dots, while the doublevelocity flow was composed of two velocities (46 and 12 deg/see). Finally, the multi-velocity flow contained more than ten velocity components ranging between 46 and 12 deg/sec (mean velocity, 23 deg/see).Each type of optic flow was presented four times to provide a large number of slow-phase eye movements for each subject and condition. Subjects, data recordingand analysis.Eye movements were recorded in three head-fixed subjects (the two authors and one naive subject). Average velocity and velocity distributionof slow-phaseeye movementswere computed for each subject and each condition,across the two trials for the same optic flowconditionto give a large total number of slow-phaseeye movements. Anrr&ses 1< -------... I I \ //’ 2 h -I-I I I Ir 10 60 B al — w— II double-velocity ——.triple-velocity w— .—...-. multi-velocity *— ml— 5 o— o2— m— s—— cu— AM /’ /’ /’ /’/ / /’ y I I\ I \ \ n 2 10 60 Slow-phase velocity [degkec] FIGURE 8. Distributions of the percentage of occurrence of slowphasevelocity (averagedacross subjects), as a function of SP velocity. (A) When observers are presented with single-velocity flows, the distribution is centred on the optical velocity (12, 23 or 46 deglsec). (B) When observers are presented with double-velocity flows containingvelocities of 12 and 46 degkec, the distribution is centred on the slowestvelocity (12 deg/see).The same pattern is observedwith a triple-velocity flow containing velocities of 12, 23 and 46 deg/sec. Finally, when observers are presented with a multi-velocity flow with velocities rangingfrom 12 to 46 degkec, the distributionis centred on the average optical velocity (23 deg.kec). were performed in a similar way to Experiment 1. Subjects were simply asked to stare at the display. Results As previously demonstrated,the distributionsof slowphaseeye velocity when subjects faced a single-velocity flow were centred on the velocity of the display [Fig. 8@)J. Slow~phaseeyevekmitydistributionselicited by a 1638 D. R. MESTRE and G. S. MASSON motion parallax display are plotted in Fig. 8(B). For the multi-velocity flow, the occurrences of slow-phase eye velocity followed a unimodal distributioncentred around the mean velocity of the optic flow (23 deghec). On the same graph are plotted the occurrencesof slow-phaseeye velocity evoked by a double- or a triple-velocity flow. The peaks of both distributionsare centred on the slowest velocity in the motion field (12 deghec). GENERAL DISCUSSION During relative motion between an observerand a rigid three-dimensional environment, different optical velocities occur in the same or neighboring parts of the visual field, due to variable distancesof objects from the observer. One consequence is that, contrary to what happenswith rotationalmotion of the observer, no single eye movement can offset the global image motion. The purpose of the experiments reported here was to investigate how the oculomotor system solves the problem of selecting one motion signal in a motion parallax flow field. Several previously published studies have demonstrated that additional depth cues, such as disparity, may help the oculomotor system to single out the motion signal related to the plane of fixation(Howard & Simpson, 1989; Busettini et al., 1996). However, as suggestedby Busettiniet al. (1996), subjectsmay wish to select a plane of attentiondifferentfrom the currentplane of fixation. Such behaviour requires the processing of differential motion. In the present series of experiments, we studied oculomotor behaviour when human subjects were presented with translational optic flows containing motion parallax. We were careful to avoid segmentation cues such as the partitioning of the visual field into central and peripheral areas (Abadi & Pascal, 1991; Gellman et al., 1990).Motion transparencyallowed us to present visual stimuli specifying a tri-dimensional structure consisting of several “depth planes” specified by several optical velocities. Segregation of motion and oculomotorcontrol When subjects are presented with transparent displays containingmotion parallax, they tend to track the slowest plane of motion. Such spontaneous behaviour was observed only when subjects were instructed to pay attention to the global scene. This result suggests that a pre-attentional segregation process is involved in the control of visual stabilization. By default, the slowest velocity output of the segregation mechanism is then selected to drive optokineticeye movements.Such a bias might reflect a low speed dominance in the velocity tuning of underlying visual processes. It might also be that, in our conditions, the lower speed also defined the most robustmotion signalcontainedin the display(due to refresh rate for instance). In other words, our results might simply reflect a particular instance of the response of the ocular system to the most robust signal in a given display. However, Niemann et al. (1994) reported a velocity dominance of OKN evoked by transparentstimuli. Their resultsrevealed that velocity dominancewas tuned to the slowestvelocity in their display,apart from one condition (visual velocities of 6 and 12 deg/sec in the display). In Experiment 1, we also found that the velocity dominance of OKN was systematically adjusted to the slowest velocity, except for the combinationof 5, 11 and 21 deg/ see, where the average slow-phasevelocity was between 5 and 11 degbec. Niemann et al. (1994) argued that this velocity dominance (around 12 degk+ecin their conditions) might reflect a “general function of the optokinetic system”.Withoutdiscardingtheir suggestion,we suggest a more functional explanation, which remains partial with regard to the available data. The role of the optokinetic response is to reduce the retinal slip of a movingvisual environment.The optimumchoice in a 3D environment might then be to stabilize the slowest moving elements, since this strategy will maintain the direction of motion of faster elements, thus avoiding shearingmotion (motion signalswith opposite signs). By stabilizingthe zone of slowestmotion in the optical flow, the optokineticsystemmight contributeto the perception of heading during translationalself-motion, as the focus of radial outflow is, by definition, motionless and correspondsto the instantaneousdirection of self-motion (Warren & Hannon, 1988). Aside from this “preference” for the slowest plane of motion observed with long-lastingstimulation,we found two situations where tracking eye movements were controlled by the average velocity in the flow field: during the initiation of the ocular response (up to 400 msec after stimulus onset) with any type of flow and during steady-state OKN when more than 10 different velocities were displayed simultaneously. We suggest that such an averaging process reveals two constraints of the pre-attentive differential motion processing enabling the segregation and, consequently, the selectionof a given motion in a complex flow field: a slow temporalbuild-up and a limited ability to segregate and simultaneouslyrepresent multiple motion signals. Motion averaging for tracking eye movements is reminiscent of depth-averagingfor vergence eye movements when two disparitiesare simultaneouslypresented in a random dot stereogram (Mallet & Arndt, 1992), or direction averaging for saccadic eye movements toward multiple targets (Findlay, 1982;Findlay & Harris, 1993). These averaging processes support a population coding scheme hypothesis. Several neurophysiological data recorded in monkeys’ area MT and MST support the idea of integrationof motion over short reaction times in the absence of segmentation cues (Bradley et al., 1995; Snowden et al., 1991; Qian & Andersen, 1994; Recanzone & Wurtz, 1994). Motion averaging may underlie the behavioral data presented here for the initiation of the ocular response, which depends, in primates, on the integrity of the same areas (Dtirsteler& Wurtz, 1988). Further studies are required to determinethe properties of the spatio-temporalmotion integrationinvolved in the initiation of tracking. Noteworthy, no data are currently EYE MOVEMENTSAND MOTION PARALLAX available concerning the spatial and temporal windows over which the oculomotor system integrates motion to drive eye movements. Psychophysical results suggest that to discriminate the direction of global motion, the visual system integrates visual motion over long periods of time (u to 450 msec) and over large spatial areas (up to 60 degJ) (Watamaniuk & Sekuler, 1992). Moreover, Watamaniuk and Duchon (1992) also demonstratedthat, when presented with short duration stimuli (between 250 and 450 msec) containing several velocities, human subjects tend to perceive the average velocity. This suggeststhat the human visual systemis able to achievea spatio-temporal integration of different velocity signals in the flow field. By investigating the properties of the initial phase of tracking, we may be able to describe the characteristics of the spatio-temporal integration of visual motion in the oculomotor system and its relationship to the spatio-temporalintegrationof visualmotion in perception. Similarities between perceptual judgments and oculomotor processes Qualitative similarities between the control of eye movements in motion parallax flow field and perceptual judgments suggest, furthermore, that similar computations, including segregation and segmentationprocesses, underlie perceptual judgments and oculomotor behaviour. One similarity between structure-from-motion(SFM) perception and the control of eye movements is the long duration of the evolutionof both processesover time. We found that, whereas the initial phase of OKN (before SFM perception) is controlled by the average velocity in the flow field, eye velocity of the subsequentslow phases (after SFM perception) decreases and stabilizes around the slowest velocity present in an optic flow, specifying two or three planes in depth. In the present study, depth perception was characterized by a long discrimination time. Accordingly, all previously published psychophysical results on tri-dimensional structure-from-motion perception emphasized the very long decision times usually observed (e.g. Braunstein & Andersen, 1981; Siegel & Andersen, 1990; Hildreth et al., 1990; Treue et al., 1991). These results indicate that the human visual system requires an extended period of time to reach an accurate perception of structure from motion. Thus, a brief observationof a moving pattern sometimesyields a perceived structure which is flatter than the “true” structure of the projected object (Hildreth et al., 1990). An incremental rigidity scheme using either position (Unman, 1984)or velocity (Grzywacz & Hildreth, 1987) information has been suggested to explain this slow temporal build-up of perception. The present results confirm that the identification of a tri-dimensional structure based on relative motion requires an extended time window, which contrasts with the short time window (about 80–100 msec) over which single image motion is measured (McKee & Welch, 1985). A comparison between the time-course of modifica- 1639 tions in tracking eye velocity and the build-up of perception might provide further insight about the nature of segregation and integration processes and about their temporaldynamics.No attempt to correlate the change in the oculomotorbehaviourand the perceptualbuild-uphas been previously reported. For instance, Ringach et al. (1996) studied vergence responses to the kinetic depth effect after the subject had reached a “steady-state” representationof the surface. From the present study, we suggest that, by looking at the relative time-course of oculomotor behaviour and perception, we may understand how high-level processes, such as the representation of a tri-dimensional structure, are involved in the control of tracking eye movements. The relationship between the time-constantof the slow-phaseeye velocity decrease and the perceptualbuild-up must be further and more precisely investigated,using the present approach. Secondly, we demonstrated that, when more than ten different motion vectors were displayed simultaneously in the flow field, the distribution of the slow-phase velocity during steady-state OKN was centred on the average velocity of the flow field. This “average” ocular responsemay be related to the limited ability of the visual system to sort out and represent several motion signals simultaneously, over the same spatial area. Supporting this hypothesis are previous results showing that human subjects cannot discriminatemore than three transparent surfaces when more than three different velocities are displayed in random dot displays (Andersen, 1989). Andersen suggested that this inability cannot be explained by the reduced dot density of each surface or by the smaller velocity increment between two “adjacent” surfaces, as the number of surfaces is increased. On the contrary, he argued that the fact that subjects could not discriminatemore than three planes might be the result of a limitation in the number of channels for information processingin both the temporal and spatial domains.The present results suggest that similar limitations exist for both oculomotor behaviour and perception. Further studies are needed to demonstrate whether such similarities are due to low-levelmotionchannelsshared by both oculomotor behaviour and 3D perception or to a direct control of perception over the oculomotor responses to complex flow fields. Additional similarities may be found between the dependencyof ocular followingresponseson dot lifetime found in the present study and previously published data on differential motion processing. Computational and psychophysical studies have already suggested that dot lifetime in a random dot display is an important constraint for the perception of structure-from-motion (SFM) (Siegel & Andersen, 1988, 1990). Treue et al. (1991) demonstrated that subjects cannot discriminate between structureand no-structurebelow a point lifetime of w 60 msec. Peak performance was reached at a point lifetime of w 125 msec. In Experiment 2, we did not precisely measure the effect of dot lifetime on depth perception. Nevertheless, for dot lifetimes of 85 msec a dramatic reduction of SFM perception was reported by 1640 D. IL MESTRE and G. S. MASSON the subjects. Similarly, changes in oculomotorbehaviour were observed, namely a significant decrease in slowphase eye velocity. Such effects of dot lifetime can also be accountedfor by an increasein the velocitynoise level in the display. In contrast, no significant changes were observed between “infinite” and 160 msec dot lifetime. Moreover, by manipulating dot lifetime, we demonstrated that the observed eye movements consisted of tracking of global motion signals and not voluntary tracking of single elements in the motion field. In summary,the present results suggestthat the control of reflexive eye movements is correlated with the subject’s perception of depth from motion. More precisely, integration and segmentation of global visual motion into distinct motion fields seem to play a key role in both processes. Nevertheless,we did not demonstrate that the visual stabilization of gaze in a motion parallax flow is controlled by the perception of depth-frommotion. To do so, it would be necessary to show that a change in eye movements related to depth, such as vergence eye movements,can be caused by the transition of tracking eye movements from one motion field to another. Recently, Ringach et al. (1996) demonstrated that the perception of a rotating three-dimensional structure from motion is sufficient to elicit vergence eye movement. It is clear from their study that high-level percepts and mechanisms can drive oculomotor behaviour which is directly related to the three-dimensional structure of the environment. Modulations of the optokinetic response by attentional processes Visual attention is one mechanism that enables us to distinguish important objects or spatial locations from less important ones. It is known that properties of the optokinetic response vary according to the subject’s attitudeto the task (ter Braak, 1957;Pola & Wyatt, 1993). In our study, subjects had to stare at either the whole motion pattern (Experiment 1) or at an actively selected motion-defineddepth plane (Experiment3). This enabled us to study the effect of attentionalprocesseson the same display and the same basic oculomotor behaviour. The present results demonstrate that when attention is allocated to the global pattern, the distribution of slowphase eye velocity is centred on the slowest velocity and not on the average velocity of the optical flow field. However, when attention is paid to one of the motion fields, distributionprofiles are centred on the velocity of the actively selected field, within velocity upper-limits similar to those observed when subjects stare at a single velocity flow. The results support the assumption that cognitiveinput does not affect the process of segregation of the flow field into its velocity components,but rather contributes to the selection of which component will be tracked. Thus, we suggest that the optokineticresponseis driven by motion parallax sensitive neuronal inputs, which segregate the global velocity field. Thereafter, the optokinetic response spontaneously selects the slowest velocity in the optical flow, determining a spontaneous gaze stabilizationstrategy in a translationaloptical flow. Only the active selection of a given depth plane is cognitively oriented from the perception of structure from motion. Ferrera & Lisberger (1995) have suggested such a network where attention only biases the selection process by which one object is attributed as the target to be pursued. Such a network implies that the different objects in the field have been independently and preattentively represented, prior to selection. Recently, Treue and Maunsell (1996) provided some insight about the putative mechanisms for such attentional biases, by demonstrating that, early in the motion pathway hierarchy, as in area MT, an attended stimulus takes greater control of cells’ responses than does the unattended stimulus. In summary, we demonstrated that optokinetic eye movementsdepend,in complex opticalflowfields,on the segmentationof visual motion into distinctmotion fields. Further research will focus on the causal links between the perception of depth and the control of conjugate and disconjugate eye movements. Precise comparative measurements of the time-course of perceptual judgments and oculomotor behaviour might also provide some insight into the temporal aspects of visual motion processing. REFERENCES Abadi, R. V. & Pascal, E. (1991).The effects of simultaneouscentral and peripheral field motion on the optokinetic response. Vision Research, 31, 2219–2225. Andersen,G. J. (1989).Perceptionof three-dimensionalstructure from optic flowwithout locally smoothvelocity. Journal of Experimental Psychology: Human Perception and Performance, 15, 363–371. Baker, C. L. & Braddick,O. J. (1982).Does segregation of differently moving areas depend on relative or absolute displacement? Vision Research, 22, 851-856. Bradley, D. C., Qian, N. & Andersen, R. A. (1995). Integration of motionand stereopsis in middle temporal cortical area of macaques. Nature, 373, 609-611. Braunstein,M. L. (1966).Sensitivityof the observerto transformations of the visual field. Journal of Experimental Psychology, 14, 582– 590. Braunstein, M. L. & Andersen, G. J. (1981). Velocity gradients and relative depth perception. Perception and Psychophysics, 72, 683– 687. Braunstein,M. L. & Title, A. J. (1988).The observer-relativevelocity field as the basis for effective motion parallax. Journal of Experimental Psychology: Human Perception and Performance, 14, 582–590. Busettini, C., Masson, G. S. & Miles, F. A. (1996). A role for stereoscopic depth cues in the rapid visual stabilisation of the eyes. Nature, 380, 342-345. Busettini, C., Miles, F. A. & Schwarz, U. (1991). Ocular responses to translation and their dependenceon viewing distance. II. Motion of the scene. Journal of Neurophysiology, 66, 865–878. Collewijn,H. & Tamminga,E. P. (1984).Human smoothand saccadic eye movementsduring voluntarypursuit of different target motions on different backgrounds. Journal of Physiolo~ (London), 351, 217–250. Diirsteler, M. R. & Wurtz, R. H. (1988). Pursuit and optokinetic deficits following chemical lesions of cortical areas MT and MST. Journal of Neurophysiology, 60, 940–965. Ferrera, V. P. & Lisberger, S. G. (1995).Attention and target selection for smooth pursuit eye movements. Journal of Neuroscience, 15, 7472–7484. EYE MOVEMENTSAND MOTION PARALLAX Findlay, J. M. (1982). Global visual processing for saccadic eye movements. Vision Research, 22, 1033–1045. Findlay, J. M. & Harris, L. R. (1993). Horizontal saccades to dichoptically presented targets of differing disparities. Vision Research, 33, 1001-1010. Gellman, R. S., Carl, J. R. & Miles, F. A. (1990).Short latency ocularfollowing responses in man. Visual Neuroscience, 5, 107-122. Gibson, J. J. (1954). The visual perception of objective motion and subjective movement. Psychological Review, 61, 304314. Gibson, J. J., Olum, P. & Rosenblatt, F. (1955). Parallax and perspective during aircraft landings. American Journal Of Psychology, 68, 372-385. Grzywacz, N. M. & Hildreth, E. C. (1987). Incremental rigidity scheme for recovering structure from motion: position-basedversus velocity-based formulations. Journal of the Optical Socie~ of America A, 4, 503–518. Hildreth, E. C., Grzywacz, N. M., Adelson, E. H. & Inada, V. K. (1990).The perceptual build-upof three-dimensionalstructure from motion. Perception and Psychophysics, 48, 19–36. Honrubia, V., Downey, W. L., Mitchell, D. P. & Ward, P. H. (1968). Experimentalstudies of optokineticnystagmus.II. Normal humans. Acta Otorhinolaryngology, 37, 65–73. Howard,1.P. & Simpson,W. S. (1989).Humanoptokineticnystagmus is linked to the stereoscopic system. Experimental Brain Research, 78, 309–314. Kersten, D., Biilthoff, H. H., Schwartz, B. L. & Kurtz, K. J. (1992). Interaction between transparency and structure from motion. Neural Computation, 4, 573–589. Kowler,E., van der Steen, J., Tamminga,E. P. & Collewijn,H. (1984). Voluntary selection of the target for smooth eye movement in the presence of superimposed,full-field stationary and moving stimuli. Vision Research, 24, 1789–1798. Mallet, H. A. & Arndt, P. A. (1992). Disparity-evokedvergence is directed towards average depth. Investigative Ophthalmology and Visual Sciences, Suppl., 33, 707. Masson, G., Proteau, L. & Mestre, D. R. (1995). Effects of stationary and moving textured backgrounds on the visuo-oculo-manual tracking. Vision Research, 35, 837–852. McKee, S. & Welch, L. (1985). Sequential recruitment in the discriminationof velocity.Journal of the Optical Society ofAmerica A, 2, 503-518. Miles, F. A., Kawano, K. & Optican, L. M. (1986). Short-latency ocular following responses of monkey. I. Dependence on spatiotemporal properties of tbe visual input. Journal ofNeurophysiology, 56, 1321-1354. 1641 Pola, J. & Wyatt, H. (1993). The role of attention and cognitive processes. In Miles, F. A. & Wallman,J. (Eds), Visual motion and its role in the stabilisation of gaze (pp.371–392).New York: Elsevier. Qian, N. & Andersen, R. A. (1994).Transparent motion perception as detection of unbalanced motion signals. II. Physiology.Journal of Neuroscience, 14, 7367–7380. Recanzone, G. H. & Wurtz, R. H. (1994). Responsesof MT and MST in macaquemonkeysto objects movingin two directions. Society for Neuroscience Abstracts, 20, 324-325. Ringach, D. L., Hawken, M. J. & Shapley, R. (1996). Binocular eye movementscaused by the perception of three-dimensionalstructure from motion. Vision Research, 36, 1479–1492. Rogers, B. & Graham, M. (1979). Motion parallax as an independent cue for depth perception. Perception, 8, 125–134. Siegel, R. M. & Andersen, R. A. (1988). Perception of threedimensional structure from visual motion in monkey and humans. Nature, 331, 259–261. Siegel, R. M. & Andersen, R. A. (1990). The perception of structure from visual motion in monkey and man. Journal of Cognitive Neuroscience, 2, 306–318. Snowden,R. J., Treue, S., Erickson, R. G. & Andersen, R. A. (1991). The response of area MT and V1 neurons to transparent motion. Journal of Neuroscience, 11, 2768–2785. ter Braak, J. W. G. (1957). “Ambivalent” optokinetic stimulation. Folia Psychiatric 60, 131–135. Neurologia and Neurochirurgica Neerlandica, Treue, S. & Maunsell,J. H. R. (1996).Attentionalmodulationof visual motionprocessingin cortical areas MT and MST. Nature, 382, 539– 541. Treue, S., Husain, M. & Andersen, R. A. (1991).Humanperceptionof structure from motion. Vision Research, 31, 59–75. Unman, S. (1984).Maximizingrigidity: the incrementalrecoveryof 3D structure from rigid and nonrigid motion. Perception, 13, 255– 274. Warren, W. H. & Hannon, D. J. (1988). Direction of self-motion is perceived from optical flow. Nature, 336, 162-163. Watamaniuk,S. N. J. & Duchon,A. (1992).The humanvisual system averages speed information. Vision Research, 32, 931–941. Watamaniuk, S. N. J. & Sekuler, R. (1992). Temporal and spatial integration in dynamic random-dot stimuli. Vision Research, 32, 2341–2347. Yee, R. D., Daniels, S. A., Jones, O. W., Baloh, R. W. & Honrubia,V. (1983). Effects of an optokinetic background on pursuit eye movements. Investigative Ophthalmology and Visual Science, 24, 1115–1122. Niemann, U., Ilg, U. J. & Hoffman, K. P. (1994). Eye movements elicited by transparent stimuli. Experimental Brain Research, 98, 31&322. Poggio, G. F. & Talbot, W. H. (1981). Mechanisms of static and dynamicstereopsis in foveal cortex of the rhesus monkey .Journal of Physiology, 315, 469492. Acknowledgements—This research was supportedby a grant from the Minist&e de la Recherche (G. Masson), CNRS and University AixMarseille. The authorswish to thank the reviewers and Cheryl FrenckMestre for significant help in improvingthe manuscript.