12 The demand to provide greater flexibility in surface- geometry measurement of free-form object... more 12 The demand to provide greater flexibility in surface- geometry measurement of free-form objects, has led to range sensors which are permitted continuous free motion. However, most devices employ mechanical linkages instrumented with position sensors, or optical or magnetic tracking sensors to determine the position and orientation of the moving range- sensor head. These additional sensors limit the working volume, impede the free movement of the range sensor, or restrict the material allowed in the work environment. This paper presents a method of surface-geometry measurement by a laser-camera range sensor head, which is permitted continuous motion in space, without the need to track its position and orientation by additional sensors.
The ability to rapidly detect moving objects while dynamically exploring a work environment is an... more The ability to rapidly detect moving objects while dynamically exploring a work environment is an essential characteristic of any active vision system. However, many of the proposed computer vision paradigms are unable to efficiently deal with the complexities of real world situations because they employ algorithms that attempt to accurately reconstruct structure- from-motion. An alternative view is to employ algorithms that only compute the minimal amount of information necessary to solve the task at hand. One method of qualitatively detecting independently moving objects by a moving camera (or observer) is based on the notion that the projected velocity of any point on a spherical image is constrained to lie on a one-dimensional locus in a local 2-D velocity space. The velocities along this locus, called a constraint ray, correspond to the rotational and translational motion of the observer. If the observer motion is known a priori, then any object moving independently through the rigid 3- D environment will exhibit a projected velocity that does not fall on this locus. As a result, the independently moving object can be detected using a clustering algorithm. In this paper, a hybrid neural network architecture is proposed for discriminating between flow velocities that are caused by camera movement and by object motion. The computing architecture is essentially a two stage process. In the first stage, a self-organizing neural network is used to learn the constraint parameters associated with typical observer movements by moving the camera apparatus through a stationary environment. Once the observer movements have been adequately learned by the self-organizing neural network, the corresponding synaptic weight values are used to program a modified radial basis function (RBF) network. During the second stage, the RBF network architecture acts as a constraint region classifier by employing clustering strategies to label incomplete motion field information (i.e. the velocity component that is parallel to the spatial gradient).
Accurate measurement and thorough documentation of excavated artifacts are the essential tasks of... more Accurate measurement and thorough documentation of excavated artifacts are the essential tasks of archaeological fieldwork. The on-site recording and long-term preservation of fragile evidence can be improved using 3D spatial data acquisition and computer-aided modeling technologies. Once the artifact is digitized and geometry created in a virtual environment, the scientist can manipulate the pieces in a virtual reality environment to develop a "realistic" reconstruction of the object without physically handling or gluing the fragments. The ARCHAEO-SCAN system is a flexible, affordable 3D coordinate data acquisition and geometric modeling system for acquiring surface and shape information of small to medium sized artifacts and bone fragments. The shape measurement system is being developed to enable the field archaeologist to manually sweep the non-contact sensor head across the relic or artifact surface. A series of unique data acquisition, processing, registration and surface reconstruction algorithms are then used to integrate 3D coordinate information from multiple views into a single reference frame. A novel technique for automatically creating a hexahedral mesh of the recovered fragments is presented. The 3D model acquisition system is designed to operate from a standard laptop with minimal additional hardware and proprietary software support. The captured shape data can be pre-processed and displayed on site, stored digitally on a CD, or transmitted via the Internet to the researcher's home institution.
A dynamic neural network, called the positive-negative (PN) neural processor, with individual neu... more A dynamic neural network, called the positive-negative (PN) neural processor, with individual neural computing units that exhibit multiple hysteresis phenomena is proposed as a plausible mechanism for the replication of certain aspects of short-term visual memory. The basic premise of the neural network model is that the cortical nervous tissue is fundamentally two-dimensional in structure. Short-term visual memory results from the immense feedback amongst the radially and laterally distributed subpopulations in the two-dimensional layer. The basic computing unit for describing the computational operations is, therefore, the neural activity generated by a particular positive or negative influencing subpopulation. STVM is defined as states of activity that persist following the removal of a visual stimulus. Once stabilized, the PN neural processor response remains unperturbed by a weak or familiar stimulus.<<ETX>>
The synergistic use of data acquired from difference sensors will enable autonomous manufacturing... more The synergistic use of data acquired from difference sensors will enable autonomous manufacturing equipment to make faster and more intelligent decisions about the current status of the workspace. Multisensor data fusion deals with mathematical and statistical issues arising from the combination of different sources of sensory information into a single representational format. A fundamental problem in data fusion is associating the data captured by one sensor with that from another sensor or the same sensor at a different point in time. This paper describes a non- statistical unsupervised hierarchical clustering algorithm used to associate the complementary feature vectors extracted from different data sets. Each level in the hierarchy consists of one or more self-organizing feature maps that contain a small number of cluster units based on the combined feature set derived from the original data. The unsupervised learning algorithm ensures that 'similar' feature vectors will be assigned to cluster units that lie in close spatial proximity in the feature map. If the sum- of-square error for the feature vectors associated with a cluster unit is greater than a predefined tolerance, then those vectors are used to create another feature map at the next level of the hierarchy. This growing procedure enables the feature set to control the number of cluster units generated. The hierarchical structure provides an efficient mechanism to deal with uncertainties in correct classification. Experimental studies are present din order to illustrate the robustness of this technique.
This paper presents a technique for reconstructing smooth closed Bezier surfaces from coordinate ... more This paper presents a technique for reconstructing smooth closed Bezier surfaces from coordinate measurements based on a Bernstein Basis Function (BBF) network. While various neural networks, such as the backpropagation network and radial basis function networks, have been effective in functional approximation and surface fitting these neural networks produce system dependent solutions that are not easily transferable to commercially available design software. The BBF network has an advantage over other networks by directly employing the same Bernstein polynomial basis functions that are used in describing Bezier surfaces. The BBF network is capable of implementing a close approximation to any continuous nonlinear mapping by forming a linear combination of nonlinear Bernstein polynomial basis functions. Changing the number of basis neurons in the network architecture is equivalent to modifying the degree of the Bernstein polynomials. Complex smooth surfaces can be reconstructed by using several simultaneously updated networks, each corresponding to a separate surface patch. A smooth transition between adjacent Bezier surface patches can be achieved by imposing additional positional C0 and tangential C1 continuity constraints on the weights during the adaptation process. Once adapted, the final weights of the networks correspond to the control points of the Bezier surface, and can therefore be used directly in commercial CAD software packages that utilize parametric modelers.
Multiple off-the-shelf cameras can be configured to simultaneously provide a large variety of par... more Multiple off-the-shelf cameras can be configured to simultaneously provide a large variety of part features that are impossible to capture with a single CCD camera or range scanner. One unsolved problem is using several cameras for passive shape recognition is that of multi-view registration. Registration is the process of associating the feature vectors extracted from the image captured by one camera view with that from another view. This paper describes an unsupervised clustering algorithm used to associate redundant and complementary features extracted from different views of a 3D object for part identification and inspection. The unsupervised learning algorithm ensures that 'similar' feature vectors will be assigned to cluster units that lie in close spatial proximity in a 3D feature map. The technique reduces the dimensionality of the input by exploiting hidden redundancies in the training data. During the inspection phase, novel features activate a number of cluster nits that have weights similar to the applied training data. During the inspection phase, novel features activate a number of cluster units that have weights similar to the applied training input. If the sum- of-square error between the input and weights of the cluster unit with the strongest response is greater than a predefined tolerance, then the part is rejected. A simulation study is presented to illustrate how the proposed multi-sensor fusion technique can be applied to identifying parts for inspection.
A neural network approach that automatically maps measured 2D image coordinates to 3D object coor... more A neural network approach that automatically maps measured 2D image coordinates to 3D object coordinates for shape reconstruction is described. The appropriately trained radial-basis function network eliminates the need for rigorous calibration procedures. The training and test data are obtained by capturing successive images of the intersection points between a projected light line and horizontal strips on a calibration bar. Once trained, the 3D object space coordinates that correspond to an illuminated pixel in the image plane is determined from the neural network. In addition, the generalization capabilities of the neural network enable the intermediate points to be interpolated. An experimental study is presented in order to demonstrate the effectiveness of this approach to 3D measurement and reconstruction.
The determination of point correspondences between range images is used in computer vision for ra... more The determination of point correspondences between range images is used in computer vision for range image registration and object recognition. The use of a spin image as a feature for matching has had considerable success in object recognition. However, in registration, refinement by iterative methods has been required. This paper present a method of determining the surface geometry in a local region surrounding the point. The technique is developed for range images which have little movement between viewpoints, and which consists of only several profiles each. The method involves fitting surface patches to the surfaces of the two successive views, creating spin-image features at a few points of each patch in one view, and determining the best match of features on the previous reference view using a localized interpolating search. The sets of corresponding points of the two successive range views are then used directly to compute the registration transformation between views. This computation effectively refines the corresponding by minimizing the residual errors. The technique is demonstrated using a pair of synthetic range views, derived from a range image of an object with a free- form surface.
Registration of conventional range images form multiple viewpoints has generally relied on redund... more Registration of conventional range images form multiple viewpoints has generally relied on redundant information from 10,000-100,000 points per image. Continuous scanning by laser-camera sensors without viewpoint knowledge requires the ability to register and integrate narrow range views inclose sequence, in order to minimize redundant data acquisition, permit high acquisition speed and reduce viewpoint planning. This paper presents a method of registering and integrating narrow and spatiotemporally- dense range views, which consists of only three profiles each, without sensor pose information.
Indirect optical methods of mechanical actuation exploit the ability of high intensity light sour... more Indirect optical methods of mechanical actuation exploit the ability of high intensity light sources to generate heat, and thereby influence the thermal properties of gases, fluids or solids. Optical actuators that utilize this photo-thermal effect for creating structural displacement often produce very large power/weight ratios. This paper describes the basic concept and operation of two optically driven micro-mechanisms that use
Robotics and Computer-integrated Manufacturing, Aug 1, 2005
A novel, interactive virtual sculpting framework based upon a deformable mesh model generated by ... more A novel, interactive virtual sculpting framework based upon a deformable mesh model generated by a self-organizing feature map (SOFM) is described in this paper. The three-dimensional lattice of the SOFM maintains the relative connectivity of neighbouring nodes in the hexahedral mesh as it transforms from the initial reference geometry into the desired shape. Material and dynamic properties are incorporated into
12 The demand to provide greater flexibility in surface- geometry measurement of free-form object... more 12 The demand to provide greater flexibility in surface- geometry measurement of free-form objects, has led to range sensors which are permitted continuous free motion. However, most devices employ mechanical linkages instrumented with position sensors, or optical or magnetic tracking sensors to determine the position and orientation of the moving range- sensor head. These additional sensors limit the working volume, impede the free movement of the range sensor, or restrict the material allowed in the work environment. This paper presents a method of surface-geometry measurement by a laser-camera range sensor head, which is permitted continuous motion in space, without the need to track its position and orientation by additional sensors.
The ability to rapidly detect moving objects while dynamically exploring a work environment is an... more The ability to rapidly detect moving objects while dynamically exploring a work environment is an essential characteristic of any active vision system. However, many of the proposed computer vision paradigms are unable to efficiently deal with the complexities of real world situations because they employ algorithms that attempt to accurately reconstruct structure- from-motion. An alternative view is to employ algorithms that only compute the minimal amount of information necessary to solve the task at hand. One method of qualitatively detecting independently moving objects by a moving camera (or observer) is based on the notion that the projected velocity of any point on a spherical image is constrained to lie on a one-dimensional locus in a local 2-D velocity space. The velocities along this locus, called a constraint ray, correspond to the rotational and translational motion of the observer. If the observer motion is known a priori, then any object moving independently through the rigid 3- D environment will exhibit a projected velocity that does not fall on this locus. As a result, the independently moving object can be detected using a clustering algorithm. In this paper, a hybrid neural network architecture is proposed for discriminating between flow velocities that are caused by camera movement and by object motion. The computing architecture is essentially a two stage process. In the first stage, a self-organizing neural network is used to learn the constraint parameters associated with typical observer movements by moving the camera apparatus through a stationary environment. Once the observer movements have been adequately learned by the self-organizing neural network, the corresponding synaptic weight values are used to program a modified radial basis function (RBF) network. During the second stage, the RBF network architecture acts as a constraint region classifier by employing clustering strategies to label incomplete motion field information (i.e. the velocity component that is parallel to the spatial gradient).
Accurate measurement and thorough documentation of excavated artifacts are the essential tasks of... more Accurate measurement and thorough documentation of excavated artifacts are the essential tasks of archaeological fieldwork. The on-site recording and long-term preservation of fragile evidence can be improved using 3D spatial data acquisition and computer-aided modeling technologies. Once the artifact is digitized and geometry created in a virtual environment, the scientist can manipulate the pieces in a virtual reality environment to develop a "realistic" reconstruction of the object without physically handling or gluing the fragments. The ARCHAEO-SCAN system is a flexible, affordable 3D coordinate data acquisition and geometric modeling system for acquiring surface and shape information of small to medium sized artifacts and bone fragments. The shape measurement system is being developed to enable the field archaeologist to manually sweep the non-contact sensor head across the relic or artifact surface. A series of unique data acquisition, processing, registration and surface reconstruction algorithms are then used to integrate 3D coordinate information from multiple views into a single reference frame. A novel technique for automatically creating a hexahedral mesh of the recovered fragments is presented. The 3D model acquisition system is designed to operate from a standard laptop with minimal additional hardware and proprietary software support. The captured shape data can be pre-processed and displayed on site, stored digitally on a CD, or transmitted via the Internet to the researcher's home institution.
A dynamic neural network, called the positive-negative (PN) neural processor, with individual neu... more A dynamic neural network, called the positive-negative (PN) neural processor, with individual neural computing units that exhibit multiple hysteresis phenomena is proposed as a plausible mechanism for the replication of certain aspects of short-term visual memory. The basic premise of the neural network model is that the cortical nervous tissue is fundamentally two-dimensional in structure. Short-term visual memory results from the immense feedback amongst the radially and laterally distributed subpopulations in the two-dimensional layer. The basic computing unit for describing the computational operations is, therefore, the neural activity generated by a particular positive or negative influencing subpopulation. STVM is defined as states of activity that persist following the removal of a visual stimulus. Once stabilized, the PN neural processor response remains unperturbed by a weak or familiar stimulus.<<ETX>>
The synergistic use of data acquired from difference sensors will enable autonomous manufacturing... more The synergistic use of data acquired from difference sensors will enable autonomous manufacturing equipment to make faster and more intelligent decisions about the current status of the workspace. Multisensor data fusion deals with mathematical and statistical issues arising from the combination of different sources of sensory information into a single representational format. A fundamental problem in data fusion is associating the data captured by one sensor with that from another sensor or the same sensor at a different point in time. This paper describes a non- statistical unsupervised hierarchical clustering algorithm used to associate the complementary feature vectors extracted from different data sets. Each level in the hierarchy consists of one or more self-organizing feature maps that contain a small number of cluster units based on the combined feature set derived from the original data. The unsupervised learning algorithm ensures that 'similar' feature vectors will be assigned to cluster units that lie in close spatial proximity in the feature map. If the sum- of-square error for the feature vectors associated with a cluster unit is greater than a predefined tolerance, then those vectors are used to create another feature map at the next level of the hierarchy. This growing procedure enables the feature set to control the number of cluster units generated. The hierarchical structure provides an efficient mechanism to deal with uncertainties in correct classification. Experimental studies are present din order to illustrate the robustness of this technique.
This paper presents a technique for reconstructing smooth closed Bezier surfaces from coordinate ... more This paper presents a technique for reconstructing smooth closed Bezier surfaces from coordinate measurements based on a Bernstein Basis Function (BBF) network. While various neural networks, such as the backpropagation network and radial basis function networks, have been effective in functional approximation and surface fitting these neural networks produce system dependent solutions that are not easily transferable to commercially available design software. The BBF network has an advantage over other networks by directly employing the same Bernstein polynomial basis functions that are used in describing Bezier surfaces. The BBF network is capable of implementing a close approximation to any continuous nonlinear mapping by forming a linear combination of nonlinear Bernstein polynomial basis functions. Changing the number of basis neurons in the network architecture is equivalent to modifying the degree of the Bernstein polynomials. Complex smooth surfaces can be reconstructed by using several simultaneously updated networks, each corresponding to a separate surface patch. A smooth transition between adjacent Bezier surface patches can be achieved by imposing additional positional C0 and tangential C1 continuity constraints on the weights during the adaptation process. Once adapted, the final weights of the networks correspond to the control points of the Bezier surface, and can therefore be used directly in commercial CAD software packages that utilize parametric modelers.
Multiple off-the-shelf cameras can be configured to simultaneously provide a large variety of par... more Multiple off-the-shelf cameras can be configured to simultaneously provide a large variety of part features that are impossible to capture with a single CCD camera or range scanner. One unsolved problem is using several cameras for passive shape recognition is that of multi-view registration. Registration is the process of associating the feature vectors extracted from the image captured by one camera view with that from another view. This paper describes an unsupervised clustering algorithm used to associate redundant and complementary features extracted from different views of a 3D object for part identification and inspection. The unsupervised learning algorithm ensures that 'similar' feature vectors will be assigned to cluster units that lie in close spatial proximity in a 3D feature map. The technique reduces the dimensionality of the input by exploiting hidden redundancies in the training data. During the inspection phase, novel features activate a number of cluster nits that have weights similar to the applied training data. During the inspection phase, novel features activate a number of cluster units that have weights similar to the applied training input. If the sum- of-square error between the input and weights of the cluster unit with the strongest response is greater than a predefined tolerance, then the part is rejected. A simulation study is presented to illustrate how the proposed multi-sensor fusion technique can be applied to identifying parts for inspection.
A neural network approach that automatically maps measured 2D image coordinates to 3D object coor... more A neural network approach that automatically maps measured 2D image coordinates to 3D object coordinates for shape reconstruction is described. The appropriately trained radial-basis function network eliminates the need for rigorous calibration procedures. The training and test data are obtained by capturing successive images of the intersection points between a projected light line and horizontal strips on a calibration bar. Once trained, the 3D object space coordinates that correspond to an illuminated pixel in the image plane is determined from the neural network. In addition, the generalization capabilities of the neural network enable the intermediate points to be interpolated. An experimental study is presented in order to demonstrate the effectiveness of this approach to 3D measurement and reconstruction.
The determination of point correspondences between range images is used in computer vision for ra... more The determination of point correspondences between range images is used in computer vision for range image registration and object recognition. The use of a spin image as a feature for matching has had considerable success in object recognition. However, in registration, refinement by iterative methods has been required. This paper present a method of determining the surface geometry in a local region surrounding the point. The technique is developed for range images which have little movement between viewpoints, and which consists of only several profiles each. The method involves fitting surface patches to the surfaces of the two successive views, creating spin-image features at a few points of each patch in one view, and determining the best match of features on the previous reference view using a localized interpolating search. The sets of corresponding points of the two successive range views are then used directly to compute the registration transformation between views. This computation effectively refines the corresponding by minimizing the residual errors. The technique is demonstrated using a pair of synthetic range views, derived from a range image of an object with a free- form surface.
Registration of conventional range images form multiple viewpoints has generally relied on redund... more Registration of conventional range images form multiple viewpoints has generally relied on redundant information from 10,000-100,000 points per image. Continuous scanning by laser-camera sensors without viewpoint knowledge requires the ability to register and integrate narrow range views inclose sequence, in order to minimize redundant data acquisition, permit high acquisition speed and reduce viewpoint planning. This paper presents a method of registering and integrating narrow and spatiotemporally- dense range views, which consists of only three profiles each, without sensor pose information.
Indirect optical methods of mechanical actuation exploit the ability of high intensity light sour... more Indirect optical methods of mechanical actuation exploit the ability of high intensity light sources to generate heat, and thereby influence the thermal properties of gases, fluids or solids. Optical actuators that utilize this photo-thermal effect for creating structural displacement often produce very large power/weight ratios. This paper describes the basic concept and operation of two optically driven micro-mechanisms that use
Robotics and Computer-integrated Manufacturing, Aug 1, 2005
A novel, interactive virtual sculpting framework based upon a deformable mesh model generated by ... more A novel, interactive virtual sculpting framework based upon a deformable mesh model generated by a self-organizing feature map (SOFM) is described in this paper. The three-dimensional lattice of the SOFM maintains the relative connectivity of neighbouring nodes in the hexahedral mesh as it transforms from the initial reference geometry into the desired shape. Material and dynamic properties are incorporated into
Uploads
Papers by George Knopf