Abnormal Video Content Searching in a Multi-camera Surveillance System

In a traditional multi-camera surveillance system, it’s hard to find the routes of the suspect objects, and search for those video clips related to the suspect objects from the surveillance database. In this paper, we present a framework for spatial relationship construction, abnormal event detection and video content searching for visual surveillance applications. This system can automatically detect the abnormal events from monitoring areas, and select the representative key frame(s) from the video clips as an index, then store the color features of the suspect objects into the surveillance database. A graph model has been defined to coordinate the tracking of objects between multiple views, so that the surveillance system can check the route of objects whether go into a critical path or not. A variety of spatio-temporal query functions can be provided by using this spatial graph model. To achieve the content-based video object searching, a kernel-based approach is employed as a similarity measure between the color distribution of the suspect object and target candidates in the surveillance database.