Abstract
In traditional security surveillance systems, only CCD cameras and storage devices are equipped for the operators to monitor the surveillance environment. Thus, the watchmen have to handle lots of thorny tasks to take care of the security problem. It is generally considered that integrate automatic alarm mechanism into traditional surveillance system will contribute to the system reliability and save lots of labor works as well. In this thesis, a novel approach was proposed to find suspicious people or objects within crowded environment. We formulate the detection system with the help of intrinsic-image-based background modeling, motion filtering, and alarm event validation. The intrinsic images which derive from consecutive frames were utilized to disclose the suspicious region within observed scenes. And the block-based motion filter incorporated with stable image updating strategy was applied to filter out the interference issue caused by large stream of moving pedestrians. Also, a block-based alarm method based on different examined period was employed to threshold the foreground objects within consecutive intrinsic images. The proposed method has been testified within various real-world video sequences and demonstrates its efficiency to disclose suspicious objects under crowded environment.