Based On Frame Difference And Background Difference Between Video Moving Object Detection Method Research

Video moving object detection is a key part in video analysis, it is the basic part for video analysis, and the test result will directly affect subsequent tracking and identification. This article research around two kinds of classic video moving target detection methods-the frame difference method and the background difference method.The frame difference method will compare two video images in different time under static background, which can reflect the moving result of object; the background difference method will extract the background image of video sequence first, and then compare the background image with original images to obtain target area. This paper puts forward an improved three frame differential method and a new background extraction algorithm, which based on research about the image threshold selection problem of the frame difference method and the background extraction problem of the background difference method.The improved three frame differential method is a moving object detection method which based on three frame difference and threshold segmentation; it is include two parts, namely three frame difference and threshold segmentation. In Original three frame difference method, segmentation threshold of the moving target area is difficult to select. The improved frame differential method is an organic combination between the original three frame difference method and the adaptive threshold segmentation algorithm, which can avoid artificial selection threshold with experience so as to effectively solve the problem. On adaptive threshold segmentation algorithm, this paper proposes an improved Otsu algorithm-an Otsu threshold video sequence image segmentation algorithm that based on genetic simulated annealing algorithm. The method has advantage on global optimization by combination with simulated annealing algorithm and genetic algorithm to optimize the threshold solving process of Otsu algorithm, and consider the video image sequence correlation between frames, and selection range of the initial population has been effectively reduced. Experiments prove that the improved threshold segmentation method is better than Otsu algorithm on the precision and efficiency, but cannot prompt poor result of Otsu segmentation algorithm when the target is too small; In the fusion of the three frame difference method and the improved threshold segmentation algorithm, detection effect is better than the original one in video moving object detection, but at the expense of the real time.This paper proposes a new background extraction algorithm to solve the image threshold selection problem of the frame difference method, namely visual background extraction algorithm on video sequence. Single frame image information in the Video sequences can be seen as superposition of background redundant information and forecast target information, so background information can be obtained through frames extraction and equalization according to residual spectrum theory. First,extract a certain number of video frames; then, make two-dimensional Fourier transform; then, get logarithmic spectrum and make equilibrium to filter forecast target information; at last, background image of video sequence can obtained by inverse discrete Fourier transform. Experiments prove that the method can accurately extract the background of the video sequences, but cannot adapt to the case which background changes greatly because of lack on background update mechanism.