Moving Object Detection In Video Sequence Based On Multi-Spectrum Image Fusion

Detection and tracking of the moving object is the main content in computer vision and image coding, and also is one of the most challenging subjects in computer vision in recent years. It has a wider application in many areas including robot navigation, intelligent surveillance systems, medical image analysis and video image compression. And the moving object detection is the foundation of object tracking, transportation monitoring, behavioral analysis, and so on. However, due to the moving object detection susceptible to changes in the background, clouds, light changes, shadows, velocity and other factors which caused the poor test results, it is very difficult to achieve the moving object detection accurately.This paper mainly studies the moving object detection of multi-spectral image sequences under static scenes. First, introduction is present about three kind of object detection method used commonly:optical flow method, time-domain finite difference method, the background difference method, and studies show that the background difference method is best suited to static scenes of multi-object detection. Second, the principle and steps of background difference method in moving object detection and existing problems are intrduced, including four background difference methods:basic background model, single gauss model, multiple gauss model and LOTS (Lehigh Omni-directional Tracking System). LOTS use two backgrounds, two thresholds and the combination of the background difference with object segmentation to reduce noises and improve the detection effect. Third, due to the visible image and infrared image have much complementary information, this paper employs the fusion of the visible image and infrared image in pixel level, feature level and decision level respectively to improve the effect of moving object detection in video. Last, the platform ViPER of object detection and tracking evaluation on video is studied, and experiments under single-spectral and multi-spectral video via the ViPER show that the method LOTS has the best effect in several background difference methods, but result of the detection with image fusion is better than that of LOTS alone.