Analyzing Tracklets for the Detection of Abnormal Crowd Behavior

Crowd behavior analysis is still an open problem in computer vision which has recently attracted more interest from vision communities. One of the main challenges is to detect abnormalities in densely crowded environments both in the space and in time domains. This implies to isolate the frames where abnormalities occur (we refer to it as frame level) and to localize within these frames, the area that generated the abnormalities (we refer to it as pixel level). The major challenge in abnormality detection is that there is no clear definition of abnormalities as they are basically context dependent and can be defined as outliers of normal distributions. With this widely accepted definition, the existing approaches for detecting abnormalities in crowd are mainly classified into three categories: i) object-based method, ii) holistic approach, and iii) hybrid methods. Figure 1. A) Shibuya crossing (japan). b) mecca (saudi arabia). c) tracklets extracted from UCSD dataset.