Research On Multi-Scene Moving Target Tracking Algorithm

As an important branch in the field of computer vision,target tracking has attracted many scholars research at home and abroad in recent years.Target tracking algorithm has been developed rapidly and has been widely used in many fields.With the continuous exploration and discovery of researchers,classical algorithms such as particle filter,correlation filter and Meanshift algorithm have been developed in the field of target tracking.Among them,CAMshift tracking algorithm proposed on the basis of Meanshift tracking algorithm is a classical algorithm in the field of single target tracking.It has attracted wide attention of relevant researchers due to its small amount of calculation,good real-time performance and insensitive to the change of target scale.However,when the moving target is in the background interference scene or occlusion scene,the accuracy of CAMshift algorithm will be seriously affected.This paper studies the CAMshift tracking algorithm,and proposes relevant improvement methods for tracking problems in background interference scene and occlusion scene.Through experimental analysis,the improved algorithm proposed in this paper can realize real-time and accurate tracking in multi-scene moving target tracking.The main research work of this paper is as follows:1.The frame image in the video is denoised,and the image is transformed from RGB space to HSV space that is more suitable for processing.The principle,mathematical model,tracking process,advantages and disadvantages of CAMshift tracking algorithm are analyzed and studied.2.Aiming at the problems of CAMshift tracking algorithm in the background interference scene,an improved CAMshift tracking algorithm based on foreground detection and feature matching is proposed.Firstly,the ViBe foreground detection algorithm is used to detect the moving target in the video,which improves the shortcomings of the traditional tracking algorithm that needs to manually frame the moving target.Aiming at the ghost phenomenon that may occur in the ViBe algorithm,the threshold of the second discriminant is calculated by the maximum between-class variance method,and the detection results of the ViBe algorithm are second discriminant,which improves the accuracy of the foreground detection.Aiming at the problem that the tracking algorithm may lose the target under the background interference,the similarity between the tracking results and the target model is calculated by Bhattacharyya coefficient,and the similarity is used to judge whether the target is lost.In the case of target loss,the AKAZE feature matching algorithm is used to reposition the tracking target,which improves the accuracy of the tracking algorithm in the background interference scene.3.Aiming at the problems of the CAMshift tracking algorithm in the occluded scene,this paper improves on the shortcomings of the accuracy of the tracking algorithm when the moving target is in the occluded scene based on the above research.Firstly,the interference scene of the moving target is judged by the similarity between the tracking results and the target model and the accuracy of feature matching,and the template update and when update are determined according to different interference scenes.When the interference scene is judged as the occlusion scene,the unscented Kalman filter prediction mechanism is introduced to predict the position of the moving target,which effectively solves the shortcomings of the traditional CAMshift algorithm that cannot accurately track the target in the occlusion scene.