Research On Pedestrian Detection And Recognition Technology Based On Binocular Vision

In recent years,pedestrian detection and recognition system based on binocular vision is a research hotspot in the field of computer vision.It has been widely used in the fields of intelligent driving,automatic guided transport vehicle and home service robot.Stereo matching algorithm and pedestrian detection technology which based on binocular vision are key technologies for pedestrian detection and recognition systems.At present,the stereo matching algorithm mostly adopts a regional-based local stereo matching algorithm which has small calculation amount and high real-time performance,but there are problems such as expansion and blur at the target edge in the disparity map;The change of body posture and posture leads to the weak detection effect of pedestrian detection algorithm and the detection accuracy is not high.Therefore,the main work is as follows:1)The edge feature of the reference image is extracted by SLIC super-pixel segmentation.The pixel points are classified according to the edge features and then the support windows of different sizes are allocated.The parallax filtering is performed on the image to be matched according to the allocated window,and finally an accurate disparity map is obtained.The experimental results show that the classification of pixel points refines the support window size of the local stereo algorithm which effectively solves the problem of edge expansion and blurring of the disparity map,and improves the accuracy of the stereo matching algorithm in the vicinity of the depth discontinuity.The error rate was reduced by 16.4%.2)The head and shoulder model is taken as the research object of pedestrian detection,a region based on adaptive threshold segmentation is used to obtain the region of interest on the disparity map,and the region to be detected is obtained by combining the region constraint method to effectively remove the parallax.The background information of the figure improves the detection efficiency.The background information of the disparity map improves the detection efficiency;The HOG features are extracted on the disparity map of the area to be detected,and the samples are trained and classified in using the SVM.The experimental results show that the head-shoulder HOG pedestrian detection method based on disparity map which effectively solves the problem of weak pedestrian detection effect and high error rate due to the variable posture and posture of pedestrians.The detection rate of this algorithm can reach 97.4%.