Research On Moving Object Detection Based On SIFT Vector Field

Object detection is an important part of Computer Vision research, which directly affects the follow-on object identification and tracking, analysis and understanding of the scene. In this paper, based on SIFT feature matching algorithm and the comprehensive reference of the idea of the optical flow, motion compensation, frame difference and so on, we put forward a moving object detection algorithm based on SIFT vector field.The optical flow method is a moving object detection algorithm, which is under the assumption that the brightness is never changing and do operation on the image pixels to make a moving vector field. First of all, by the revelation of the optical flow algorithm, we can use the SIFT feature matching algorithm to extract the features of the two adjacent frames in the video, and do operation by using the matching SIFT features to create a SIFT vector field. It's more stable and faster than the optical flow algorithm, for SIFT feature is more stable and the number of the SIFT feature matches is less than the number of the points in the image. Then, by doing clustering analysis for the vectors in the SIFT vector field, we can get the classes of the SIFT vector field, each of which represents a background or a object. To distinguish between the background and objects, we can calculate the inner distance of classes for dispersion analysis. Maximum dispersion is the background, the rest are objects. Next, by drawing on the idea of the frame difference algorithm, each object is being compensation according to the value of the vector in the SIFT vector filed, and this makes the two images overlap in the object area, then make difference between them and make threshholding for the difference image, it makes the object area in the image turns to black, and other areas turn to white. For more processing, we should make opposition on the image, and make the object area turns to white. Finally, do morphological processing to eliminate noise, and then the contours of the object can be obtained.Experiments show that the algorithm not only can detect moving objects under a static background, but also can detect moving objects under a dynamic background, with good stability, calculation speed, and detection accuracy and so on. In addition, this algorithm can handle occlusion, eliminate camera shake caused by voluntary movement, and still it suited for object detection under complex background.