Abnormal events' detection in crowded scenes

In this paper, two new methods are developed in order to detect and track unexpected events in scenes. The process of detecting people may face some difficulties due to poor contrast, noise and the small size of the defects. For this purpose,the perfect knowledge of the geometry of these defects is an essential step in assessing the quality of detection. First, we collected statistical models of the element for each individual for time tracking of different people using the technique of Gaussian mixture model (GMM). Then we improved this method to detect and track the crowd(IGMM). Thereafter, we adopted two methods: the differential method of Lucas and Kanade(LK) and the method of optical flow estimation of Horn Schunck(HS) for optical flow representation. Then, we proposed a novel descriptor, named the Distribution of Magnitude of Optical Flow (DMOF) for anomalous events' detection in the surveillance video. This descriptor represents an algorithm whose aim is to accelerate the action of abnormal events' detection based on a local adjustment of the velocity field by manipulating the light intensity.