Studies On Moving Object Segmentation Based On Symmetrical Differencing Algorithm In Video Sequences

The researches of segmentation of moving object are important for computer vision field .This dissertation describers our investigation on some of computer vision. Research fruits mainly include segmentation algorithm of moving objects from sequential images.Now traditional moving object detection and segmentation of moving object algorithm mainly include temporal differencing method, background subtraction ,based on optical flow etc. Background subtraction using the differencing of currently image and background image to detect moving area, it is sensitive to changing of dynamic scene, the key is to set up background and dynamic renovate. temporal differencing algorithm is obtain the moving area through the difference by using sequence two frames and three frames or multitudinous frames in sequence image. It is adapt to dynamic scene, but defect is object don't move so fast, and it couldn't distill whole character of object, existing inanition or the being short of edge. optical flow algorithm is first estimate moving field , based on moving field to segmentation moving object . The virtue of the method is that it could segmentation moving object when the vidicon moving. But the defect is the working of calculate is big, reliability is not so good.This dissertation aiming at existing defect of traditional segmentation by temporal differencing method, such as when the surface of object has no enough grain ,it is easy to come into being inanition and being short of edge. We suggest a new moving object segmentation algorithm base on "three frames symmetry differencing method" .the algorithm is to multiply the result of three symmetry ,boost the second pixel dot ,to increase the different of goal and noise of background ,adopt threshold, detect of edge, morphologic operator, connectedness operator etc. ,and combine background renovate algorithm being handled. Experiment results show that the algorithm is the moving object avoid appear inanition and being short of edge, finally we obtain a full object of being segmentation.The principium of traditional temporal differencing algorithm is suppose vidicon is immobile, if the differencing of sequence frames is not zero, then it is because noise and changing of moving object . In order to reduce vein background, the displacement of two frames is small. To non-rigid body, because it move slow, the displacement is so small that detect result is not so good, so usually by using symmetry differencing algorithm to segmentation moving object.Experiments result show that when non-rigid body moving to view, if it's vein is lack, by using temporal differencing algorithm to segmentation, usually appear inanition. This question would infection next work , such as object distinguish . Researches show that the cause of that is vein is lack , so that morphologic operator couldn't detect different object. To resolve this question, the text combine background subtraction and symmetry differencing method. It describes current frame image and saved background image difference subtraction, if pixel difference is bigger than threshold, then determinant the pixel dot is moving object dot. On based differencing moving object detect, the key is to obtain background image. And background model is often renovate in time.The text method describe :At first, collection first frame image make initialization background, then, obtain segmentation of moving object , based on symmetry differencing existent inanition or the being short of edge. Cover the moving object surface , make label with connectedness. Renovate label field pixel with different. This method calculate with quick speed, and avoid infection with changing of light. Through three frames symmetrical differencing segmentation moving field . Then by using background subtraction and symmetry differencing algorithm obtain pixel differencing , the result and symmetrical differencing addition, to have many moving object .Experiments result show that, the method could obtain whole moving object , effect is all right. Experiment hardware circumstance is Pentium 4, 2.0G CPU. Software is using Matlab 7.1 to program.The dissertation suggest a new moving object segmentation algorithm based on symmetry differencing algorithm . The new algorithm show that effect is best in segmentation of moving object in video image. Avoid inanition and the being short of edge , to make basing of mode recognition. The improved method mainly based on symmetry , combined the virtue of background subtraction algorithm . So the new method is quick speed.