Video is becoming an integral part in various applications such as entertainment, education, medicine, databases, security, and even wireless applications. A video surveillance system is supposed to interpret what is happening in the dynamic scene and to raise warning if some abnormal events occur. The video processing framework of a surveillance system includes different modules such as enhancement, segmentation, feature extraction, tracking, and interpretation. The output from these modules can be 1) binary; 2) textual description; and /or 3) video signal. MPEG is a collection of standards for the compression and description of such outputs. It provide a wide family of industry standard for multimedia (video and audio) compression and description. The objective of this project is to develop a system which uses video processing techniques to build a reliable door surveillance system.
To achieve this objective, students will build a model of an entrance (e.g., to a home) and use IP cameras to monitor it. The cameras are connected to a computer which signals an alarm if the house is broken into and identifies which window or door is used. It also detects if there is any suspicious movement in the house or the area around it at times when the occupants are away. The system also issues an alarm if kids approach an area which is potentially dangerous to them. Students will build on an available video surveillance system to design an advanced real-time system to 1) capture digital video signal via a mobile (wireless) camera; 2) process the digital video captured; and 3) transmit extracted data over IP. The project will include the extraction of MPEG-7 features (visual descriptors) such as edge histogram and dominant color. The extraction code must run in real-time. The extracted features are used to construct an MPEG-7 stream which describes the video objects. The stream is archived in a database and used to query the surveillance video. For example, to ask the system to show the images of objects whose dominant color is blue and so on.
The system will be composed of a server (to manages communication between the workstations), video workstations (to capture and process video), and control workstations (to access the system). When objects are segmented and tracked, an alert signal together with the compressed images and description of the moving objects will be sent to the control station. The video transmission part will follow over IP using the real-time streaming protocol (RTSP). In this project, students will have access to different pre-defined modules (C++ library implementation) for enhancement, segmentation, tracking, MPEG-7 compatible feature descriptor design, and event identification.
The project include a design of a Linux system to integrate different video processing modules. The input to this system will be real time video scenes from any fixed camera through a standard network or web. The system also needs to integrate functionalities from Matlab to evaluate overall performance. Thus the proposed server interface will work as a benchmark for qualitative analysis for a successful surveillance video application. The demonstration must include real time (30f/s) video processing and quality analysis of the different modules installed in an operational environment (e.g., high-way, bus station, subway, airport terminal). |