Moving Object Segmentation In The Video Captured By A Panning Camera

With the extensive application of multimedia, more and more things shown byvideo, the research about video is increasingly becoming a hot topic. Segmentation, asone of the fundamental problems in digital video processing, which provides the basisfor other processing and application, is more important.In the past, videos of segmentation almost are captured by a fixed camera, andsegmentation technology can be divided into three basic methods, i.e. BackgroundSubtraction, Frame Difference and Optical Flow. Background Subtraction, with thebackground to segment the foreground, is widely used, and many methods improve theprocedure used to update the background model. Frame Difference does not define thebackground model, but rather uses the neighboring frame to segment, then gets theboundary of the foreground. So the result of Frame Difference needs to fill the blackpart. Optical Flow assumes that the intensity of moving object remains constant along amotion trajectory. Because of relative motion between the foreground and thebackground, Optical Flow can segment the foreground and background throughcalculating motion vector between two successive frames. For the moving camera, thesemethods can’t segment effectively the foreground from the background.With the movement of the camera, the background also changes, but the differencebetween the background and the foreground, which is caused by relative motion, doesnot change. According to the basic segmentation methods above, the proposedalgorithm in this paper extracts the motion vectors from the video, and calculates themeans and variance of the pixels in horizontal and vertical directions. Using the meansand variance, the horizontal and vertical displacements, which are the basis ofbackground motion vector, can be got by comparing the neighboring frames. Then forthe previous frame, the proposed algorithm gets the motion compensation imageaccording to the global motion estimation by background motion vector. Adopting theidea of Background and Frame Difference, the contour of foreground objects can beextracted by removing the background of the current frame and gets the foreground ofsegmentation. According to the experimental results, the proposed method can extractthe moving foreground in real-time, effectively. For the complex background, thismethod can get the better result of segmentation.