Moving Objects Detection And Tracking In Video Surveillance

Moving object detection and tracking has been a key issue for the intelligent video analysis. It refers to detect and track the moving object in the video surveillance without human intervention. The process would contribute to analyzing the action of the moving object, describe the action of the object, judge the abnormal phenomena in the scenes and ultimately deal with the abnormalities.Regard to the difficulties of video surveillance research, the experts and scholars at home and abroad constantly propose new solutions and applications. Bases on the great achievements of the experts and scholars, the main tasks of this thesis on moving object detection and tracking in video surveillance are as followings:(1) We analyzes the causes and the characteristics of the shadows and concludes the advantages and disadvantages of the various shadow elimination algorithms. Meanwhile, the shadow elimination algorithm based on the combination of the illumination invariant image and the image difference is put forward. The proposal of the new shadow elimination algorithm is to solve the problem that the moving object is extracted incompletely with the illumination invariant image. The cause for the phenomenon is that the illumination invariant image is applicable in the assumed condition of the Planck light, Lambert light reflection model and each channel of the camera responding in narrow band to the spectrum. The assumed condition doesn’t exist in the natural environment.(2) We propose a new detection algorithm of moving object which combines the modified LBP(Local Binary Pattern)texture algorithm and HSV(Hue, Saturation, Value) color invariant algorithm. The improved LBP Texture Shadow Elimination is put forward when the number of LBP texture operators is overlarge. The modified HSV Color Invariant Shadow Elimination is raised as the original one is not suitable for the unstable hue-saturation. There are more advantages when the combination of the two algorithms is applied than each of the algorithms is adopted individually. The background is described with LBP and HSV. Eventually the background is applied to Gaussian Mixture Model to realize the dynamic update of the background.(3) We propose a new method which could improve the mean-shift tracking algorithm of moving object. The modified mean-shift tracking algorithm could solve the problems of inexact tracking under the circumstance of morphological change and morphological occlusion to a certain degree and this method could be applied in more fields.