Abstract
The concept of pedestrian priority has been taken gradually in recent years. However, there are still many vehicles and motorcycles caused traffic accidents due to ignoring the pedestrian right. In this work, we design and implement a human recognition system based on image processing techniques for pedestrians crosswalk. This system can be used to improve the safety level and reduce the probability of intersection accidents. We use the environment feature vectors obtained by the system to detect the zebra-crossing and find out the range of the zebra-crossing. We propose a dual-camera mechanism to maintain the detection accuracy and improve the fault tolerance of the proposed system. We design an Enhanced-Motion-HOG classification scheme to recognize pedestrians at the road intersection. To verify the feasibility and efficiency of our system, we implement a prototype system and compared the accuracy of the pedestrian detection scheme with the original HOG method and the motion detection method. Experimental results demonstrate that our scheme outperforms these schemes in terms of detection accuracy and processing time.
You have full access to this open access chapter, Download conference paper PDF
Similar content being viewed by others
Keywords
1 Introduction
Pedestrians are a powerful indicator of the social and economic health and safety of a community. The recent studies about the casualties of the accident presented that among the road users, pedestrians are the most vulnerable groups [8]. They quite need to be protected in dangerous road intersections.
In the last few years, researchers have focused on pedestrian detection technology, for enhancing pedestrian detection accuracy and simplifying pedestrian detection processing time. Many pedestrian detection algorithms have been developed [2, 10].
In this research study, we take a new approach with HOG in the past which apply suitable detect pedestrians at the intersection. We also show that analysis specifically, comprehensive about processing time and detection accuracy. The remainder of this paper is organized as follows. Section 2 describes the recent relevant research and technology. The design and architecture of the proposed system are explained in Sect. 3. Section 4 details the system prototype. Finally, Sect. 5 presents conclusions.
2 Related Work
In this section, we will introduce some related work. First, we discussed some image processing schemes about how to recognize zebra pattern. Then, we present related studies which involve the pedestrian detection based on HOG.
Many researchers have already studied the zebra-crossing region detection. Se [9] proposed a method to detect zebra-crossing which can be considered as a group of consecutive edges. Khaliluzzaman and Deb [5] proposed a zebra-crossing detection framework based on unique geometrical features of zebra-crossing. Ahmetovic et al. [1] presented the ZebraRecognizer algorithm that rectifies the ground plane hence removing the projection distortion of the extracted features.
Beiping et al. [2] presented a real-time Human detection algorithm based on HOG features and SVM architecture. He et al. [4] improved the HOG + SVM pedestrian detector with the semantic regions of interest obtained from a fully-convolutional neural network. Kim et al. [6] proposed the HOG-UDP algorithm that included a dimension reduction functionality based on the manifold.
3 System Architecture
This section details the architecture of the pedestrian crosswalk recognition system. Figure 1 shows an overview of the system. In this system, Camera is used to collect the frame images from real-time video. After that, we have to detect the target region which includes the crossing region and the waiting region. To identify the crossing region, we need to recognize zebra-crossings and find the boundaries of them. The waiting region is detected following the crossing region. These regions will not the same depending on each intersection environment.
After target region detection, we will perform to recognize pedestrian in there. We design a set of Motion-HOG classifier with mobility. The mobility rate and pedestrian’s characteristics as the screening factor for pedestrian recognition.
The camera is vulnerable to wind, earthquake, rain and the impact of various environmental factors shaking, change the location, so the pedestrian waiting line and the area is not static. To obtain the crossing region and the waiting region accurately, we observed that all the intersections are recognizable features zebra-crossing so that we can see the zebra line as a picture calibration point to slightly correct the screen deviation. The location of the zebra-crossing is mapped to the waiting region. After obtaining the environmental parameters, we use these parameters as the control values of the zebra-detection algorithm. We set up a dual-lens camera on both sides of the zebra crossing. The system will combine the detection results from both sides, we can fix the detection errors caused by long distance.
The Motion-HOG classifier is shown in Fig. 2. First, when the nth frame in the webcam stream was inputted to the system, the background subtraction phase first establishes the background model through the mean-background method [4, 9] and then extracts the foreground object dataset.
In the Motion detection phase, we use object tracking method to find out if there is any movement object belong to the dataset. And the velocity of speedobject would be calculated. When the velocity speedobject of the objectj is lower than the speed threshold value speedthreshold, the objectj will be delivered to the HOG detection phase and analyze whether it is Pedestrian objects and output the results. We use two threads in our method to handle HOG detection and Motion detection functions and detect all the motion objects at the same time, and then detect all HOG features simultaneously. It can reduce the detection time.
4 System Prototype
The prototype system is implemented to provide an adaptive zebra-crossing finding function and pedestrians detection function in the intersection.
We use Raspberry Pi 3 (RPi3) Module B [11] as our system device which is equipped with a ARM Cortex-A53 CPU 1.2 GHz, 64-Bit, Quad-Core and 1 GB RAM, as shown in Fig. 3. The Python language is used to implement the system interface and OpenCV library was chosen to implement our main function in OS system Raspbian.
After we receive the area coordinate, we started to detect the pedestrians who are waiting in the area via our Motion-Hog method. We implement the dual camera view detection to ensure our system could decrease the detection error rate and there is a signal light in the upper right corner to indicate whether there has pedestrian or not. We selected the Kaixuan Road in Taichung, Taiwan as the study site for case design with setting up the dual camera as shown in Fig. 4.
5 Conclusions
In this paper, we design and implement a human recognition system which aimed to detect pedestrians crosswalk. The zebra-crossing recognizer adapts the variation of environmental illumination. We can obtain the crossing region and waiting region. After that, we can know whether pedestrians are passing or are going to pass the zebra-crossing by our proposed Motion-HOG classifier. In the future, we will develop the pedestrian crossing warning system based on the pedestrian detection results in this research. Furthermore, we will implement the detection of the vehicle to expand the ability of our system.
References
Ahmetovic, D., Bernareggi, C., Gerino, A., Mascetti, S.: ZebraRecognizer: efficient and precise localization of pedestrian crossings. In: 2014 22nd International Conference on Pattern Recognition, Stockholm, Sweden, pp. 2566–2571 (2014)
Beiping, H., Wen, Z.: Fast human detection using motion detection and histogram of oriented gradients. J. Comput. 6(8), 1597–1604 (2011)
Gradinescu, V., Gorgorin, C., Iftode, L., Diaconescu, R., Cristea, V.: Adaptive traffic lights using car-to-car communication. In: Vehicular Technology Conference (VTC 2007), Dublin, Ireland, pp. 21–25 (2007)
He, M., Luo, H., Chang, Z., Hui, B.: Pedestrian detection with semantic regions of interest. Sensors 17(11), 1–16 (2017)
Khaliluzzaman, M., Deb, K.: Zebra-crossing detection based on geometric feature and vertical vanishing point. In: 2016 3rd International Conference on Electrical Engineering and Information Communication Technology (ICEEICT), Dhaka, Bangladesh, pp. 1–6 (2016)
Kim, P.-K., Kim, H.-H., Kim, T.-W.: An improved pedestrian detection system that utilizes the HOG-UDP algorithm. In: Park, J., Chen, S.-C., Raymond Choo, K.-K. (eds.) MUE/FutureTech -2017. LNEE, vol. 448, pp. 218–225. Springer, Singapore (2017). https://doi.org/10.1007/978-981-10-5041-1_37
Koonce, P., Rodegerdts, L., Lee, K., Quayle, S., Beaird, S., Braud, C., Urbanik, T.: Operational and safety analysis. In: Traffic Signal Timing Manual, pp. 56–87. Federal Highway Administration, Turner-Fairbank Highway Research Center, McLean, VA, USA (2008)
Najada, H.A., Mahgoub, I.: Big vehicular traffic data mining: towards accident and congestion prevention. In: International Wireless Communications and Mobile Computing Conference (IWCMC), Paphos, Cyprus, pp. 256–261 (2016)
Se, S.: Zebra-crossing detection for the partially sighted. In: IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2000, Hilton Head Island, SC, USA, pp. 211–217 (2000)
Sladojević, S., Anderla, A., Ćulibrk, D., Stefanović, D., Lalić, B.: Integer arithmetic approximation of the HoG algorithm used for pedestrian detection. Comput. Sci. Inf. Syst. 14(2), 329–346 (2017)
Raspberry Pi. https://www.raspberrypi.org/
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2018 Springer International Publishing AG, part of Springer Nature
About this paper
Cite this paper
Dow, CR., Lee, LH., Huy, N.H., Wang, KC. (2018). A Human Recognition System for Pedestrian Crosswalk. In: Stephanidis, C. (eds) HCI International 2018 – Posters' Extended Abstracts. HCI 2018. Communications in Computer and Information Science, vol 852. Springer, Cham. https://doi.org/10.1007/978-3-319-92285-0_60
Download citation
DOI: https://doi.org/10.1007/978-3-319-92285-0_60
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-92284-3
Online ISBN: 978-3-319-92285-0
eBook Packages: Computer ScienceComputer Science (R0)