Research On Key Technologies Of Target Tracking In Wild Scenes Based On Deep Reinforcement Learning

As one of the most challenging subjects in the field of computer vision,target tracking tasks are continuously researched and optimized by researchers.The main purpose of this task is to mark the position and size of the target to be tracked in the first frame of a given video sequence.The research work and innovations proposed in this paper are as follows:First,this paper proposes a deep reinforcement learning model based on doubledelay deep decision-making for single target tracking,which uses a double-delay deep decision algorithm to further optimize the Actor-Critic model.We use two Critic networks to jointly predict the bounding box confidence,obtain the smaller predicted value as a label,and then update the network parameters to accelerate the convergence of the loss function to obtain the most accurate result.Through comprehensive experiments on the benchmark test libraries OTB-2013,OTB-2015 and VOT2016,the results show that the algorithm proposed in this paper has excellent performance in terms of accuracy,robustness and real-time performance.Since the benchmark test libraries OTB and OVT lack data in the field environment,this paper puts forward the relevant data sets in the field in the harsh environment.However,the target tracking algorithm of double-delay deep decisionmaking is tested on such data sets,and the results have the characteristics of high accuracy and low overlap rate.In response to this problem,we improved the dualdelay depth decision algorithm by inputting the original bounding box,the enlarged bounding box and the reduced bounding box into the Actor model at the same time to obtain three moving target bounding boxes.The image in the bounding box is matched with the image in the target bounding box of the previous frame to obtain the best bounding box.The algorithm we proposed can solve the situation of sudden size change and large size change,and effectively improve the robustness of the algorithm.Finally,this paper designs and implements a target tracking system,which can be used to track targets in real time,where the target is selected by the user interactively.After the tracking is over,the system will show the accuracy and overlap rate during the tracking process.