Group Sports Video Surveillance

Intelligentized crowd surveillance technology is an important research sub-field in intelligent video surveillance system. As a key technology of intelligent surveillance, it is of great value in a large number of applications such as crowd management, public space design, virtual environments simulate, visual surveillance, intelligent environments analysis, etc.In this paper, moving crowd is divided into two catalogs in terms of sparseness, namely, sparse crowd and dense crowd. The crowd density estimation and the crowd motion estimation are investigated, based on which a surveillance software is designed and implemented.A scheme interweaved with object detection and tracking is used to grasp and track individual object for sparse crowd counting. First, non-parametric model for background subtraction is used to detect moving objects. Certain appearance and motion features are extracted to find a model to represent a person. Then, people are tracked by Kalman filter and feature matching. Finally, the number of pedestrian is provided through object association.In consideration of the fact that there may exist certain relationship between crowd density and feature, feature-based regression was applied to count number of dense crowd. Area of blobs, amount of harris and KLT-feature points, number of contours, perimeter of contour, ratio between the contour perimeter and area, total Canny edge pixels and fractal dimension are employed, while applying perspective and occlusion correction to weight each objective pixels. Then crowd count was estimated by a regression function, e.g. linear regression, non-linear regression, stepwise regression analysis, multivariate linear regression.Two algorithms were proposed to make crowd motion estimation. One is feature points optical flow computation for estimation of crowd motion, the other is based on motion history image. The results of these two algorithms on video data show the effectiveness in crowd motion estimation.In addition, this paper reported the development of a crowd intelligent monitoring software, which has the functions of pedestrian count, crowd density estimation and density anomaly detection, anomaly detection based on motion direction.