Research On Traffic Target Detection Method Based On Deep Learning

With the continuous advancement of urbanization,in recent years,China’s car ownership is growing rapidly,the modern road traffic environment is more and more complex,and people spend more time on traffic,which puts forward higher requirements for the traffic system.Intelligent transportation system can improve traffic efficiency and ensure traffic safety.Traffic target detection is the basic link of high-level tasks such as target tracking and behavior recognition,so it has become the basic key technology in many aspects of intelligent transportation.Among them,the traffic target detection method based on machine vision is cheap and easy to promote,It has become a research hotspot of intelligent transportation.In this paper,the traffic target detection method based on deep learning is studied,and the improvement is mainly aimed at improving the detection accuracy and reducing the performance requirements in traffic target detection.The specific research work is as follows:1.Compare all kinds of traditional target detection methods and the relatively new target detection methods based on neural network,analyze their respective characteristics,summarize the shortcomings in the application of traffic target detection and the causes of the shortcomings,through the analysis and summary of all kinds of common target detection methods,finally select YOLOv3 detection method as the basic framework of this paper.2.Analyze the problems encountered in the traffic target detection using the original YOLOv3 detection method,and analyze the causes of these problems respectively.Aiming at the inaccurate measurement of the coincidence degree between the prediction box and the ground-truth box by the complete intersection over union function in special cases,improve it,and propose a new intersection over union function PIOU(Position Intersection over Union)to improve the detection accuracy;By increasing the detection scale,the feature map is refined and the deep and shallow semantic information fusion is enhanced to enhance the small target detection ability;By adding SE(Squeeze-and-Excitation)attention mechanism module,the network pays more attention to the feature information of learning important channels;By fusing local features and global features,a new feature map is obtained to further improve the detection accuracy;The experimental results show that the proposed method can improve the detection effect and reduce the detection time.The maximum map of88.3% is achieved on KITTI dataset,which can achieve better detection effect under the condition of low hardware requirements.