Research On Improved YOLO Algorithm In Pedestrian Detection And Tracking

Pedestrian detection and tracking is an important application direction in the research of object detection and tracking.It can obtain the location and moving track of pedestrian target accurately in real time,provide important reference information for the analysis of pedestrian abnormal behavior,and is the core technology of computer vision tasks such as automatic driving.However,in real application scenarios,it is inevitable for pedestrians to block between small size and pedestrians in images or videos,which will have a great impact on the effect of pedestrian detection and tracking,and to a large extent limit the application of pedestrian detection and tracking algorithm in some real scenes.In view of the above problems,this paper mainly studies the feature fusion and completes the following research work:(1)In order to solve the problem that features are selectively discarded and shallow information is ignored in the progressive process of network structure of YOLO algorithm.In this paper,the DenseNet network structure is studied,and the deep and shallow features are fused by the DenseNet network structure,so that the deep network structure can fully mine the visual features of the shallow layer and make full use of the accurate location information of pedestrians in the shallow features.The experimental results show that the combination of INRIA pedestrian data set and VOC pedestrian sub data set has a good effect.(2)In order to solve the problem of insufficient and incomplete pedestrian feature extraction in the actual application scene with pedestrian occlusion.In this paper,the head and body alignment model is used to enhance the pedestrian detection ability,and then the head and body alignment model is optimized by the alignment loss function to produce accurate detection results.The experimental results show that the head body alignment model can solve the problem of pedestrian occlusion and improve the accuracy of pedestrian detection.(3)Because Deep-Sort tracking algorithm is based on detection and detected by Faster-RCNN network,it can not meet the real-time requirements when using Deep-Sort algorithm to track pedestrians.In this paper,the improved YOLO algorithm and Deep-Sort algorithm are combined,and the improved YOLO algorithm is used to replace Faster-RCNN for pedestrian detection as the basis of pedestrian tracking,so that the pedestrian will not be lost under occlusion.The experimental results show that the method proposed in this paper can effectively solve the problem of pedestrian following loss and meet the real-time requirements of pedestrian tracking.