A Study Of Single Target Tracking Technology Based On Deep Learning In Complex Background

Recent years,with the maturity of Graphics Processing Unit(GPU)technology.Deep learning has made breakthroughs in the field of computer vision.A large number of tracking algorithms and detection algorithms that based on deep learning have been proposed in recent years with demonstrated success.The failure of tracking algorithms is often caused of scale change,occlusion,complex background and target disappearance,etc.In order to solve tracking failure we studied the target detection algorithm and tracking algorithm that based on deep learning.The main lines of this paper include:Firstly,this paper structure a UAV data set which is used for object detection.In order to improve the accuracy of small-scale UAV detection in dataset,this paper reference residual module make object detection algorithm(SSD)better.This paper improves an object detection model based on the residual network(ResNet50),which optimizing the depth of object detection algorithm.With the depth of detection network model increases,the feature information becomes more variety.The residual network contribute to strengthen sensitivity of output to input variations.Secondly,this paper improves an object detection algorithm based on the lightweight neural network MobileNet_V2.Some convolution modules are filled to object detection algorithm.The MobileNet_V2 is based on inverted residual structure and streamlined architecture MobileNet_V1 which uses depth-wise separable convolutions to build light weight networks,so that we can effectively reduce the number of model parameters.After that,this paper presents a tracking framework based on lightweight target detection algorithm which are mentioned in foregoing paragraph.This tracking framework is applied into video with complex background.Then a confidence APCE is introduced to the kernel correlation filtering tracking algorithm.APCE is used to judge whether the tracking is successful or not.The lightweight target detection algorithm is used to track and rectify the target.When tracking failure,the lightweight target detection algorithm will find the location of the target and make the tracking again.It significantly improved the robustness of the tracking algorithm and guaranteed the real-time performance of the tracking algorithm.This tracking algorithm is more suitable for complex situations such as motion blurring,target disappearance and fast moving speed in complex background.This paper studied the single-target UAV tracking algorithm based on deep learning in complex background.To some extent,it solved the problem of how to reduce background confusion and target disappearance in complex background,which leads to tracking failure,and improved the performance and robustness of the single-target tracking algorithm.