The Research On Foreground Detection Algorithm Based On Improved Gaussian Mixture Model

Foreground detection has been widely used in many computer vision applications, including:video monitoring, auxiliary driving, human-computer interaction, face detection and pedestrian collision prediction, remote sensing image processing, the foreign bodies in food inspection, identification of pedestrian behavior, etc.. It’s a hotspot and frontier in the field of computer vision and pattern recognition.The stand or fall of foreground detection results is directly related to the quality and practicability of the subsequent application and analysis. There are a variety of different types of foreground detection algorithms, the gaussian mixture model algorithm and its improved methods are one type of the most popular algorithms. On the basis of the detailed analysis and understanding of the principle of the gaussian mixture model, this thesis mainly studies the foreground detection algorithm based on improved gaussian mixture model under the circumstance of fixed camera. The main works are as follows:(1) Summarize the foreground detection, and introduce some commonly used methods of foreground detection:frame differential method, optical flow method and background subtraction method. At the same time, introduce some commonly used methods of the background subtraction method, and take experimental test on them, detailedly analyze and compare their respective advantages and disadvantages.(2) Take a comprehensive introduction and analysis on the traditional gaussian mixture model algorithm to learn that the considerable accuracy of gaussian mixture model is at the cost of time-consuming. At the same time, the treatment effect of noise of gaussian mixture model is very general. Aiming at the disadvantages of the gaussian mixture model, put forward a kind of foreground detection algorithm which is based on YCbCr adaptive gaussian mixture model. Firstly, the algorithm replace RGB color space with YCbCr color space, in order to get better anti-noise property. Secondly, considering the real time performance, adaptive selection strategy is used to determine the number of gaussian components of the gaussian mixture model.In the end, the gaussian components will be ordered by their different values of sorting keys, to select the background models. Experiments show that this algorithm can well deal with the influence of noise, in addition to sudden illumination changes in scene, it can obtain more accurate outline of foreground objects in the indoor and outdoor scenes in general, and has a good real-time performance.(3) According to the gaussian mixture model only use color feature to model, and it can not deal with sudden illumination changes, put forward an improved gaussian mixture model foreground detection algorithm based on the multiple features combination. This algorithm select the YUV color feature with LBP texture feature for gaussian mixture modeling to make up for the single feature is not comprehensive, and to improve the way of updating the gaussian mixture model. Experiments show that this algorithm can effectively deal with the sudden change of illumination and the dynamic background.