Pedestrian Detection And Tracking By Driverless Cars

In recent years,with the rapid development of artificial intelligence technology,driverless cars have become a research hotspot and will become the future development trend of the automotive industry.The key technology of self-driving cars is the perception of vehicles on the road environment,and use the sensed information to issue correct driving instructions to ensure the safety of driving.Pedestrians,as the most common objects in the road environment,play an indispensable role in the detection and tracking of pedestrians in driverless technology.Therefore,the study of pedestrian detection and tracking algorithms is extremely important.This paper builds two frames for pedestrian detection and pedestrian tracking.When the vehicle speed is 40 Km / h,a real-time road scene is shot on the road by a high-speed motion camera as a test sample to verify the effectiveness of the pedestrian detection and tracking algorithm.The research work in this paper will focus on two aspects: pedestrian detection and pedestrian tracking:In terms of pedestrian detection,an improved pedestrian detection algorithm based on tiny-YOLOv3 network is proposed.Use k-means clustering on the pedestrian training set to obtain the optimal candidate frame size and number;improve the network structure of tiny-YOLOv3 and improve the ability of the network to extract pedestrian features.The algorithm proposed in this paper improves the detection speed under the premise of ensuring higher detection accuracy.In terms of pedestrian tracking,a multi-pedestrian tracking algorithm based on data association is proposed.First,a tracking framework is built based on the Kalman filter algorithm;then,the IOU distance is used as the evaluation matrix,and the Hungarian matching algorithm is used for trajectory matching;a wide residual network is used to extract pedestrian depth appearance features,and a second match is used to solve pedestrian goals The target disappears due to occlusion.This algorithm shows better results than other algorithms when pedestrians are blocked.After experimental verification,the algorithm proposed in this paper has good accuracy and real-time performance for the road environment video taken in real time,and can meet the needs of unmanned vehicles to a certain extent.