Research On Image Segmentation Based On Crowd Surveillance Video

As one of the hot issues in the research areas of image processing and computer vision, video object segmentation plays a very important role in the field of video surveillance, human-computer interaction, behavior analysis, and video editing. The purpose of image segmentation is to simplify or change the manifestation of the image, making the images easier to understand and analyze. In the actual scene, due to the camera angles, shadows, as well as the mutual contact between the targets, one motion block will contain more than one person. It is a serious challenge to segment and track of each human body accurately. Aiming at the problem of crowd segmentation, a new method of segmentation and tracking of crowd objects proposed.Firstly, for the complex scene and noise that affecting segmentation issue in the fixed occasion, a new background modeling approach for moving objects detection is proposed. This model is based on classic mixture Gaussian background model. And it classifies for each pixel in Time and Space scales. Further, HSV color combined with Local Binary Pattern texture description is used to remove the shadows.Secondly, this paper introduces an improved Hough circle detection method to detect the target’s head, at the same time, the combination of the Kalman filter algorithm to achieve the target tracking.Finally, TurboPixels algorithm is used to pre-segment the foreground area into several super-pixels. Then use template matching algorithm to identify each of the initial target area, the final goal is to create a weighted graph model based on the dissimilarity of corresponding region and the degree of mismatching of a human model. Then the best segmentation boundary is defined by the optimal path.