Research On Method Of Segmentation And Tracking Of Pedestrians Of Vehicular Infrared Image Sequences

Pedestrian detection based on infrared imaging technology has important academic significance and application value.Infrared image mainly captures the heat generated by objects to form the gray-scale image,the imaging principle is simple,free from light,color,texture and smoke and so on,so it has broad application prospect in the automotive driver assistance system and intelligent monitoring system(especially at night).Far-infrared pedestrian segmentation,as the basis of far-infrared pedestrian detection and tracking,plays an important role in improving the performance of pedestrian detection and tracking.Pedestrian tracking has a broad application prospects in the improvement of pedestrian detection performance and the study of pedestrian movement.This paper focuses on the segmentation of pedestrian objects in vehicle far infrared images based on vehicle-assisted driving and the tracking of pedestrian targets.According to the principle of far infrared imaging and the characteristics of the pedestrian targets in the image,this paper makes the following research:1)A segmentation method based on road information and pixels' brightness vertical projection is proposed.According to position information of the road,the Areas of Interest(AOIs)is obtained adaptively.By vertical projection of the pixels' brightness of the AOIs region,the initial vertical position of pedestrian areas may exist.As pedestrians are limited by their own width,the vertical strips to get the screen,then we can get the most likely vertical strip pedestrians area.The good performance and anti-noise ability of the proposed method are verified by comparison with the current excellent segmentation methods.2)A segmentation method based on the combination of Mean shift algorithm and vertical projection algorithm is proposed.The segmentation method of pixels' brightness projection may divide the pedestrian into two candidate regions and be susceptible to the background objects with high brightness.The segmentation and merging of the images are carried out by Mean shift algorithm,combining the pixels' brightness vertical projection Method,the image block is filtered to obtain the final candidate region.The experimental results show that this method can reduce the number of candidate regions and the miss rate pedestrian detection.Finally,the pedestrian detection technology proves that this method of3)segmentation significantly improve the performance of pedestrian detection.4)This paper proposes a tracking algorithm based on improved Kernelized Correlation Filters(KCF).Aiming at the problem of KCF template updating and scale fixation,this paper proposes an improved method based on the characteristics of pedestrian targets in vehicle far-infrared image sequences.First,the templates are updated by weighted accumulating the templates of all the frames to make the templates more representative.Secondly,aiming at the problem of target scale change,the method of selecting multi-scale targets is proposed.And the scale of template is updated at the same time.Finally,the parallel processing of the trackers is carried out to solve the problem of multiple targets in the image.The experimental results show that this method has good performance in the infrared image sequences.