Background Modeling Methods And Their Application In Moving Objects Detection

Moving object detection, which is related to computer vision, image processing, video processing, and many other theories, is an important research focus in intelligent video surveillance under complex scene; its research has important theoretical significance and practical value.Background modeling method is a most widely used and most important method in moving object detection. Both national and international researchers have carried on widely studies and investigations about background modeling theoretically and practically, and obtained a great deal of achievements, however, there still exist many issues. Based on deeply research of the previous achievements, background modeling methods is researched deeply in this thesis. The main works and research results are summarized as follows:First, both gaussian mixture background model and non-parametric background model were studied thoroughly in this thesis, and at the same time the two methods also were compared in the theory and experiment. Hence the following conclusion are arrived: the non-parametric model is better than gaussian mixture model a little, but because of its algorithm's time complexity and space complexity are higher, the use of gaussian mixture model is more widely. In addition, shadow elimination and denoising of the two modeling methods'post-processing have been studied. Good results were gained on shadow elimination based on HSV color space; also a quick and simple method of denoising is proposed, which is to calculate the size of the area of all the potential target connected region, and then a thresholds is set up for denoising. The result of experiment proves that this method is superior to morphological operators for de-noising.A background modeling method is proposed based on CSLBP (Center-Symmetric Local Binary Patterns). CSLBP texture model inherits the advantages of LBP texture model, at the same time, CSLBP's anti-noise ability is stronger than LBP, and its computational complexity reduces significantly. Experimental results show the advancement and effectiveness of the background modeling method.