Multi-features Fusion Algorithm Research In Moving Object Detection

Moving Object detection is an important researching topic in computer vision community, which has widely application prospects in video surveillance, transportation, medical research, the military field and so on. With the scholars at home and abroad have been doing in-depth study about this theory, moving object detection technology has been rapidly developed.The paper puts the intelligent surveillance video analysis techniques as research background, aiming the true complexity of the environment, discusses the interfering factors such as the illumination changes, shadows, scene itself changes and moving background. In response to these problems, we study some commonly used moving object detection algorithms and give the comparative analysis of these theoretical principles and results and then finally sum up the characteristics of their respective. Particularly, we carry out a research into background modeling algorithm in moving object detection, discuss the theories of texture-based modeling and color-based modeling respectively, analyze the theoretical results and then draw the conclusion that the two types of algorithms are complementary. On these foundations, in this paper, we propose a multi-features fusion algorithm. Using the idea of Gaussian mixture modeling, at the pixel level, the algorithm sets up two models including color models and texture models. While at the region level, it seeks the common fields between the color model regions and the texture model regions. To do so, it can ignore the effect caused by shadows and illumination and get the real foreground. In texture model, local binary pattern is used to represent the texture feature in order to make the background can inhibit the effect of shadows. Photometric invariant color in RGB color space is used in color modeling for making background robust to shadows and illumination changes. For the update of the background weight, a "hysteresis" scheme is introduced to which can speed up the update rate.In this paper, we implement the proposed method and evaluate the performance of our method on real scenes. The experimental results prove that our method can apply to the complex scene under illumination conditions and meanwhile it can reduce external interferences, such as the light changes, moving background (branches swaying) and shadow impact. Finally, the paper compares the proposed method with Gaussian mixture modeling, by analyzing the contrast results, obtains the proposed method has strong accuracy and robustness under complex environment containing shadows and moving background.