Research On Pedestrian Detection And Tracking Technology Based On Improved YOLOX And Deepsort

With the development of science and technology,pedestrian detection and tracking technology is widely used in intelligent driving,public place monitoring,fire safety and other important fields.However,due to the large population density on the road,the existing algorithms have low detection efficiency and slow speed,resulting in poor tracking effect.Moreover,most target detection algorithms need to obtain the relevant super parameters of a priori frame size for different sample data sets,so as to obtain the candidate region of the detection target.The model training operation is complex,and the size of a priori frame directly affects the detection accuracy of the model,so it is difficult to have an appropriate algorithm to obtain accurate relevant size parameters.Therefore,this paper proposes the algorithm research of pedestrian detection and tracking based on YOLOX and Deepsort without anchor frame.The specific research is as follows:(1)Aiming at the problems of large flow of road people and slow detection speed,the backbone network of YOLOX is optimized.Firstly,in view of the large amount of model parameters and high requirements for hardware equipment in csparknet53 network,the lightweight convolution module ghost is replaced with conventional convolution,and asymmetric convolution is used to reduce the amount of operation parameters.After the improvement,the amount of network parameters is reduced by 50%,and the excellent network structure of the original model is maintained,so that the detection speed of the model is increased from 45 frames/s to 53 frames/s.(2)Aiming at the low detection accuracy of the model and the problem of missing detection of pedestrians,the CBAM module combining spatial attention and channel attention is used to enhance the semantic information.For the IOU loss function of pedestrian detection is difficult to calculate the difference between the prediction frame and the real frame,DIOU loss is used as the loss function of the model to effectively improve the accuracy of regression.The map value of the improved model reaches 95.34%,which is 0.85% higher than that of the original model.(3)Use the data to train the pedestrian target of Deepsort.The improved YOLOX model is used as the detector of Deepsort to improve the tracking speed,and the detection speed reaches 48 frames/s.The experimental results show that the improved algorithm can effectively detect and track pedestrian targets,and meet the real-time requirements,which has a certain practical value.