Deep Learning Motion Larget Automatic Tracking Camera System Research

The field of computer vision has been greatly developed by the influence of deep learning in the process of its growing development.As one of the important research directions in the field of computer vision,many mature target tracking techniques have been proposed in recent years,and influenced by the huge computing power of the Internet,target tracking techniques will have great potential for development in the field of intelligent surveillance.This topic is a research for deep learning based motion target automatic tracking camera system.Firstly,this paper conducts an in-depth research on the relevant theory of deep learning,target detection technology and target tracking technology,in order to make the target tracking algorithm more in line with the requirements of the target tracking system on the tracking success of the algorithm,this paper improves the detection part of the tracking algorithm,and then improves the tracking accuracy of the target tracking algorithm.The specific improvement strategy is as follows.The SENet channel attention module is added to the original network of the Yolov5 detector.The addition of the SENet module increases the attention of useful feature information according to the channel importance,and suppresses the less relevant feature information.Replace the maximum pooling operation in the original feature pyramid module with a soft pooling operation,which minimizes the loss of feature information during the pooling process.The GIOU in the base network is replaced with the CIOU loss function.The improved target detection model improves the average detection accuracy metric by 4.4%,verifying that the method can effectively improve the detection accuracy.After completing the improvement of the target detection model base network,the improved target detector was combined with the Deepsort target tracker to form a new target tracker.The new target tracker is experimentally validated on the MOT16 dataset,and finally,the analysis of the experimental data shows that the improved target tracker has improved in both MOTA and MOTP metrics when compared with the pre-improved target tracker.In this paper,after completing the improvement of the target tracker,the experimental platform of the automatic motion target tracking system was then built,the hardware part of the tracking system was selected and the software part of the process was designed,and the experimental verification of the motion target tracking system was carried out in an everyday life scenario.The experimental data analysis shows that the improved tracking algorithm combined with the control algorithm running on the hardware platform can automatically track the moving target better and reduce the dead angle range of the monitoring view,and the method can be applied in the field of video surveillance.