Research On Target Tracking Algorithms For Edge Computing

Thanks to the development of correlation filter and deep learning technology,target tracking algorithm has made great progress in recent years,and has been applied in many fields.The main challenges of tracking algorithms include target occlusion,scale change,illumination change and motion blurring.How to solve these problems is still the main research hotspot in this field.In addition,due to the realtime requirement of tracking task,it is difficult to deploy target tracking algorithm based on deep learning model on mobile terminal devices.In view of the above problems,this paper studies and improves the single-target tracking algorithm from two aspects of algorithm model improvement and application deployment optimization,devotes itself to the research of stable,reliable and excellent tracking algorithm.In addition,combined with the edge computing platform,from the perspective of application deployment optimization,the deployment strategies of target tracking algorithm on mobile devices are studied,and a reasonable and reliable deployment scheme is proposed.Firstly,the problem modeling and solving methods of tracking task are introduced.The background-aware correlation filter is analyzed from the perspective of model,and a temporal regularized background-aware correlation filter is proposed.By adding temporal regularization term to the filter model,that is,adding smoothing constraints to the models learned from continuous video frames,the robustness of the algorithm to occlusion distortion,illumination change and some other problems is improved.In addition,a motion model is added,and the updating method of the model is optimized to reduce the error updating of the algorithm.Experiments on OTB100 dataset show that the performance of the improved algorithm is significantly improved,and 10 of the 11 attribute tests in the dataset are dominant.In addition,the characteristics of mobile devices and the architecture of edge computing platform are analyzed in this paper.Aiming at specific application scenarios,a target tracking system based on edge computing platform is proposed,which integrates target tracking algorithm based on correlation filter and deep learning.The system offloading computing tasks to edge clouds reasonably through task segmentation module,and uses information fusion module to analyze and fuse the results.In addition,motion detection module is added to further reduce the computing pressure and power consumption of terminal nodes.Finally,a comparative experiment of different deployment strategies is carried out.The experimental results show that the deployment strategy significantly reduces the task response time compared with local computing of computing tasks,and the deployment strategy reduces the same computing task processing time compared with completely uninstalling to the edge cloud.