Research On Pedestrian Detection And Tracking Technology For Automatic Driving

In recent years,the automotive field has set off an upsurge of research on autonomous driving technology.The realization of autonomous driving technology is an important research topic in the field of artificial intelligence,and a common research hotspot of academia and industry,with great commercial value.As an important part of automatic driving technology,pedestrian detection and tracking plays an important role in the safety of automatic driving.The main content of this paper is the research of pedestrian detection and tracking algorithm.Firstly,it introduces the research status of automatic driving technology at home and abroad,and then introduces the main framework and basic principle of pedestrian detection and tracking algorithm combined with the application scenario of automatic driving.Finally,it puts forward the pedestrian detection and tracking algorithm which is suitable for the automatic driving scenario designed in this paper.In the pedestrian target detection algorithm,a lightweight pedestrian detection network is proposed.Mobilenetv3 is used as the backbone of YOLOv3 pedestrian detection algorithm to reduce the number of network model parameters and computation,and improve the detection speed of the network to meet the real-time requirements in the automatic driving scenario.The improved CIOU function is used as the loss function of Mobilenetv3 network.The Attention mechanism module adopts Coordinate Attention to enhance the network’s ability to extract local features.The feature fusion network part adopts SFPN network to further enrich the pedestrian target features.Finally,the pedestrian detection network is verified in the BDD100 K data set and City Person data set.The accuracy of the improved pedestrian detection network is further improved,the amount of model calculation is reduced,and the detection speed is also greatly improved,which verifies the feasibility of the lightweight pedestrian detection network.In the pedestrian target tracking algorithm,Deep Sort algorithm is further improved,and the extended Kalman filter is used to predict the pedestrian target.In the data association part,GIOU matching algorithm is used to realize the data association between the pedestrian target detection results and the predicted results,which further improves the accuracy of pedestrian tracking.Finally,the pedestrian tracking network was verified on the MOT20 dataset.The tracking accuracy of the improved Deep-Sort tracking network is further improved,which verifies the rationality of the improved scheme.