A Fusion Method Of Extracting Moving Object Based On Optical Flow And Level Set

As an important branch of computer vision in image processing, motiondetection is one of the theoretical and applied research focus in recent years. It iswidely used in financial, transportation, security and many other fields, playing amore and more important role in all aspects of social production and life.Motion detection, by means of the computer, is extracting the interesting movingtargets from the background in a continuous image sequences. The motion detection isimportant to deal with aspects of motion analysis and intelligent monitoring etc.Motion detection can not only obtain motion information in the image, but alsosimplify the difficulty of the subsequent steps such as motion tracking, identificationand classification.Traditionally, it is very difficult for a single motion detection algorithm alone todetect moving targets effectively and completely from complex natural scene imagesequences. Multi-method fusion, which can obtain a higher accuracy and efficiency ofmotion detection, detecting moving targets by the integration of movement targets andimage information, is attracting more and more attention by researchers. Whileimproving the accuracy, fusion method often inhibit the efficiency of motiondetection. It causes the final result with a serious distortion at the same time when thenoise exists in image sequences.A new simple but effective fusion method based on optical flow field and levelset method is proposed in this paper. Optical flow field is acquired by the adjacent twoframes of the image sequences firstly. Binary image is got from optical flow by settinga threshold. Then, the coarse moving object area can be obtained by mathematicalmorphological close operation, which is aimed at attaining the initial contour line of alevel set curve evolution method. After selecting the appropriate image color model,the precise contour is searched by the energy function for evolution to extract movingobject. Experimental results show that this method is suitable for detecting the movingobject rapidly and effectively without depending on the accuracy of optical flow fieldheavily.Finally, by the experimental results of vehicles in real traffic scene, it is verifiedthat the proposed algorithm has a certain degree of practicality and universality.