Study Of Robust Detection And Tracking Of Moving Object Under The Condition Of Complicated Scene

Intelligent visual surveillance has become the focus of the field of machine vision research in recent years. The core of the research in this area is the moving target detection, tracking and follow-up behavior recognition based on video. It deals with the issues of high-tech in many fields such as pattern recognition, image processing, artificial intelligence, automatic control and machine vision. There have broad application prospects and practical significance in the military guidance, visual navigation, security surveillance, intelligent traffic, video coding fields.This paper describes the basic theories and key technologies of moving target detection and tracking in video sequence images. It focus on techiniques of the robustness of the moving object detection, tracking and counting in complex scenarios.For moving target detection in comlex scene. First, It provides an overview and summary for the three major current detection methods(ie, background subtraction, inter-frame difference, optical flow method), and gives the corresponding experimental results. It also analyzes and summarizes the advantages and disadvantages of each method. Then it has done a detailed analysis and discussion for the moving target detection algorithm based on Gaussian mixture model. And on this basis, the algorithm has been improved:This algorithm assumes that the color information between adjacent pixels be no correlation on Gaussian mixture model, Gaussian background model of each pixel is independent in each other. We build a background model for each pixel, using thoughts of target pixel inter-frame correlation, the random noise frames independence, and combining with the time filter(Temporal Filter)algorithm focus on the direction of optical flow motion characteristics, later processing on the mathematical morphology. It has effectively solved the moving target detection problems similar to camera jitter, swaying branches, the water ripples cyclical movements conditions in scene. The experimental results show that the detection result of the algorithm is good, fast and stable.For target tracking and counting, In order to achieve real-time target tracking and stability effects. In this paper, we introduced a theory by combining Kalman Filter and Mean Shift tracking algorithm. Begining, it applies an adaptive background subtraction method to obtain prospects for the motion regions, filters the target area of interest according to pre-set threshold. Then it predicts the next moment movement area that moving target may be using Kalman Filtering algorithm. Combining with Mean Shift algorithm to calculat the feature and obtain the mass of the moving target detected for real-time tracking it is. A counting method is proposed by using frame tracking information to real-time track moving target in the video scene. It has achieved the desired results.