Research On Road Target Detection And Tracking Method Based On YOLO V

With the rapid development of the global economy and the continuous improvement of people’s living standards,cars have become the preferred means of transportation for many families.With the continuous growth of car ownership,road traffic safety problems have become more serious,autonomous driving technology as an effective means of solution,camera vision-based object detection and tracking technology is one of the key technologies.In the driving perspective,due to the small target in the image field of view,and under the congested road section,there are problems such as mutual occlusion between the targets,which makes the detection difficult and seriously affects the detection accuracy and tracking effect.In order to solve the above problems,this paper carries out research on road object detection and tracking algorithm based on driving perspective based on the existing object detection and tracking algorithm.Based on the YOLO v5 network model,this paper uses the SPPF module instead of the SPP module to improve the speed of the module to process feature maps.Introduce bottomup PANet network to enhance the feature fusion capability of the network.A new YOLO detection head has been added to enhance the learning perception of small targets;Drawing on the idea of detection head decoupling in YOLOX algorithm,the classification task and regression task of object detection are decoupled.The number of cycles of the C3 module in the Neck network was changed from 3 times to 1 time to reduce the number of layers and parameters of the network and improve the detection speed of the model.Secondly,this paper compares and analyzes the existing target frame regression loss function,and proposes an improved CIoU target box regression loss function(ICIoU)to obtain a smaller target box regression loss error.The K-means++ clustering algorithm is used to correct the initial prior box,so that the size of the newly generated initial prior box is more consistent with the target of the dataset in this paper.The Soft NMS algorithm is used to screen the target frame to improve the detection efficiency of the occluded target.The improved network structure and algorithm were experimentally analyzed using SODA10 M dataset,and it was experimentally verified that the m AP of the improved detection algorithm was improved by 8.8% compared with the YOLO v5(5.0)network and 8.6% higher than that of the YOLO v5(6.0)network.The results show that the improved detection algorithm in this paper can complete the object detection task more accurately in different road scenarios.At the same time,the performance of the improved detection algorithm in this paper is compared with other classical algorithms,and the experimental results show that the improved detection algorithm has better detection performance under the premise of meeting real-time performance.In this paper,the improved detection algorithm is embedded in the framework of DeepSORT multi-target tracking algorithm,and the apparent model is retrained to realize multi-class target tracking from the driving perspective.Experimental results show that under the KITTI tracking dataset,using the detection algorithm in this paper as the tracking detector of DeepSORT algorithm,the multi-target tracking accuracy is improved by 1.568% and the multi-target tracking accuracy is improved by 1.01% compared with YOLO v5.The improved detection algorithm in this paper can not only significantly improve the performance of road object detection,but also improve the tracking performance of road objects,which is of great significance for the realization of autonomous driving.