Design And Research Of Target Tracking System Introducing-detecting Network

Target detection and target tracking technology has always been a basic and important research direction in computer vision.With the rapid development of UAV technology,it is also very important to use UAV as an application platform to achieve target detection and target tracking.In this paper,the target tracking algorithm introduced into the detection network has been researched.First,the context-aware filter tracking algorithm is improved for the contextual background information problem,and then the related filter tracking algorithm is improved for the feature and scale issues in the tracking process.The improved algorithm above fails to deal with the problem of target occlusion and similar object interference.The target tracking algorithm introduced into the detection network is improved,and finally the UAV obstacle detection,tracking and obstacle avoidance system is built.The main work and innovations of this paper are as follows:1.For the context-aware related filter tracking algorithm,only the background information around the target is roughly processed.It is considered that the importance of the background information around the target is the same.This article improves a context-aware adaptive related filter tracking.Algorithm,the algorithm first predicts which direction the target will move next by using particle filtering,and then when training this tracking filter,the background area around the target area is used as negative samples,and the target will be next.The weight given to the background area in the moving direction is much greater than that in other non-moving directions,which improves the tracker’s ability to discriminate between the target and the background,which is verified by experimental simulation.2.Aiming at the feature problem and scale change problem in the target tracking process,this paper improves a target tracking algorithm based on multiple features and scale adaptation.Using the advantages of different features in different environments,the color histogram features and depth features are merged.At the same time,in order to solve the problem that the scale of the target has indeed changed,a scale filter is introduced to judge the target scale and select the appropriate target scale.The context-aware filter is used as the framework to improve the tracking performance of the algorithm in scenarios such as illumination changes and target scale changes,and it is verified by experimental simulations.3.Aiming at the problem of target occlusion and interference from similar objects,this paper improves a target tracking algorithm that introduces a detection network on the basis of the Siamese FC algorithm.When there is no single peak in the tracking response graph,it is considered to be in the tracking process.In this case,the target may be blocked or the target similar to the interference may occur,so the target position in the response map is re-detected through the detection network,so that the target position can be retrieved for tracking.In this way,the algorithm can track the target stably and accurately,which is verified by experimental simulation.4.Based on the above research on target detection and target tracking,a UAV obstacle detection,tracking and obstacle avoidance system based on deep learning is built.First,the Keruixiang binocular camera mounted on the drone is used to collect images,and then the distance between the obstacle and the drone is calculated through video analysis,and then the obstacle avoidance algorithm is used to calculate the desired position and the desired angular velocity and send it Give flight control to control the drone to avoid obstacles.After testing,this system can realize the UAV’s obstacle detection,tracking and ranging,and complete the obstacle avoidance flight,achieving the expected effect.