Research On Moving Object Detection And Tracking Algorithm Based On Local Feature Analysis

With the development of technology and information technology, moving object detection and tracking as a basis for intelligent video surveillance and the core in the field of computer vision has become a hot issues for researchers, its effects directly influence the work of the whole system. However, due to the complexity of the background, the natural environment and other objects outside interference, detecting and tracking moving objects of interest poses a severe challenge.This thesis on the basis of learning and training the classical moving object detection and tracking algorithm, mainly to completes the following tasks:(1) Learning, classifying and summarizing the existing classical moving object detection and tracking algorithm, focuses on the idea background subtraction algorithm, and experimenting this algorithms, analyzing the existence of their respective advantages and disadvantages.(2) In order to solve the deficiencies of Gaussian mixture model algorithm for dealing with shadows, light and noise, this thesis on the basis of it, presents a moving object detection algorithm based on local feature analysis, which uses the textures feature to create the modeling of the video background, using the histogram of the Local Binary Patterns with uniform to represent the texture features and using histogram matching and updating the relevant mechanisms to achieve real-time detection of moving objects. Comparing the results of this thesis and the frame difference method and the Gaussian mixture model, the experimental results show that the algorithm in dealing with noise and light and shadow achieves good results.(3) After the realization of the moving object detection, this thesis presents a moving object tracking algorithm, which is combined of by the Kalman filter and the texture model. Firstly, the algorithm marks the prospect of connecting region with the rectangular and calculates the rectangle area histogram of its texture, then uses the Klaman filter states information of moving objects and update forecast correction, and using texture histogram match and update the search-related goals to achieve tracking of the moving objects. Through experiments and analysis show that the algorithm in dealing with linear, Gaussian distribution is very well, which can achieve the real-time tracking, and overcome some of the light changes.