Research On Dynamic Traffic Target Tracking And Trajectory Prediction Algorithm Based On Vision

At present,through the continuous development of economic and technological level,automobiles have begun to enter millions of households,but the following traffic safety,traffic congestion and other problems were constantly invading us,and autonomous driving technology is an important way to solve the above problems in the future.At the same time,through the rapid development of computer chip computing power,artificial intelligence has also developed by leaps and bounds.It has played an important role in face recognition,target detection,image processing and other fields.Therefore,the use of artificial intelligence to realize autonomous driving systems for autonomous vehicles is mainly It consists of three parts: environment perception,path planning and decision control.Because of its low cost and rich perception information,the vision-based environment perception algorithm is based on indepth learning of related algorithms to develop vision-based dynamic traffic targets.Research on tracking and trajectory prediction algorithms.For urban traffic scenes,research has been carried out in three aspects: target detection,target tracking,and trajectory prediction.The main research contents are as follows:(1)Target detection algorithm based on deep learning.This paper compared and analyzes the difference in detection speed and accuracy between the traditional target detection algorithm based on artificial features and the target detection algorithm based on deep learning,chose to use the YOLOv3 algorithm based on deep learning,and used the video production data set captured by the driving recorder.Training,in order to improve the adaptability of the detection model,night,rain,and cloudy images were added to the data set for data enhancement.In addition,in view of the large volume of the target in the road traffic scene,the structure of the YOLOv3 network was pruned to remove the network layer for small targets,so that it had a faster detection speed.(2)Target tracking algorithm based on target detection.This chapter studied the SORT algorithm,which was a target tracking method based on Hungarian matching and Kalman filter.Based on the target state information detected by YOLOv3,Kalman filter was used to calculate the estimation frame,and Hungarian matching with the detection frame was performed to complete the data association.This algorithm had a small amount of calculation,and as long as the target detection algorithm was excellent enough,the performance of the algorithm could also meet the needs of use.(3)A trajectory prediction algorithm based on time series.This article started with the temporal and spatial information of the historical trajectory of the traffic target,choosed to use the long and short-term memory(LSTM)network with excellent time series prediction effect to process the historical trajectory of the traffic target,and trained the LSTM network by using the historical trajectory information of the target to make it It could predict the short-term future trajectory of the target based on the historical trajectory of the target,and had good accuracy.