Font Size: a A A

Research On Moving Target Tracking Based On Computer Vision

Posted on:2020-11-19Degree:MasterType:Thesis
Country:ChinaCandidate:S M HeFull Text:PDF
GTID:2428330596977371Subject:Control engineering
Abstract/Summary:PDF Full Text Request
As a product of the information society,computer vision has always been a hot research topic.Nowadays,computer vision-based tracking technology has been applied to many fields such as security monitoring,traffic monitoring,coal mine safety operations and military.Therefore,research on this technology is of great significance.This paper mainly summarizes some of the shortcomings of the detection and tracking algorithms of moving targets based on the previous research methods of scholars,and mainly improves the following aspects:Selection and improvement of target detection algorithms.Firstly,the detection performances of inter-frame difference method,optical flow method and background difference model method are compared in theory and experiment.Then the Vibe background difference model algorithm with excellent detection performance is used for theoretical analysis and experimental research.The initial frame modeling strategy based on the traditional Vibe algorithm makes the detection result easy to mix into the Ghost pixel.The Ghost region is determined by combining the frame difference method and the update factor is adjusted to achieve the purpose of quickly integrating the Ghost region into the background.In addition,the dynamic threshold update background model is used to eliminate the influence of background disturbance on the detection result,so that the Vibe algorithm can detect the moving target more accurately.Finally,the shadow part of the foreground pixel is detected in HSV space to filter out the shadow pixel misdiagnosed as the foreground,and the experiment proves that the improved Vibe algorithm has better target detection effect.Research and improvement of target tracking algorithm.Firstly,the theoretical basis and experimental effects of the two basic tracking algorithms,Meanshift and Camshift,are comprehensively analyzed.After careful measurement and comparison of the tracking characteristics,the Camshift algorithm which is robust to the target size change is selected for key research.The modeling of its dependence on a single H component is not enough to accurately distinguish the moving targets in complex scenes.This paper also uses the S and V color components to describe the target to enhance the color feature information of the target itself.In addition,this paper also combines the rotation-invariant unified LATP features and HSV color features that are robust to illumination changes and are not interfered by similar color backgrounds,forming a joint feature with stronger target discrimination and a probability density map of joint features.Maximizing the Pap s Coefficient is the premise that the iteration seeks the position most similar to the target.Enhance the performance of the algorithm for illumination changes and target tracking in similar color backgrounds;for the disadvantages of the Camshift algorithm for tracking the moving targets and nonlinear moving targets under occlusion,and to further enhance the stability of the Camshift algorithm tracking,After fully analyzing the tracking characteristics of Camshift algorithm and particle filter algorithm,a joint algorithm tracking mechanism is given,which makes full use of the advantages of the respective algorithm to achieve joint tracking of complex scene targets.In addition,the joint target tracking algorithm is combined with the improved Vibe detection algorithm to reduce unnecessary errors introduced by manually selecting the target area,and then the automatic target tracking algorithm of this paper is formed.Finally,the moving target in complex environment is tracked.Experiments verify that the automatic tracking algorithm has high tracking accuracy and the real-time tracking is better.
Keywords/Search Tags:Target tracking, ViBe, CamShift, multi-feature, particle filtering
PDF Full Text Request
Related items