A Modular Distributed Video Surveillance System Over IP Drew Ostheimer S´ ebastien Lemay Mohammed Ghazal Dennis Mayisela Aishy Amer Pierre F. Dagba Electrical and Computer Engineering, Concordia University Montr´ eal, Qu´ ebec, Canada Correspondence: amer@ece.concordia.ca Abstract We present an automated and distributed real-time video surveillance system which can be used for the detection of objects and events in a wide range of applications. Video feeds are captured from multiple sources, processed and streamed over the Internet for viewing and analysis. Components of the system can be interconnected in several manners, thus forming flexible systems. The experimental results show a system that handles multiple video feeds, running on standard computers and yielding fluid video. Several interconnected clients can view multiple feeds simultaneously, as well as the event listing. Keywords — Video, Surveillance, Internet Protocol, Mul- ticast, RTSP. 1 Introduction In response to the increase in security concerns, auto- mated real-time surveillance applications have received sig- nificant attention from the research and industrial commu- nities. The deployment of such applications has become a necessity in airports, subways, offices and even homes. Cur- rently deployed automated surveillance systems suffer from non-scalability or low frame rates due to computationally expensive algorithms. In general, a video surveillance sys- tem should have 1) affordable hardware requirements, 2) real-time environment adaptation, 3) low bandwidth con- sumption, 4) access control procedures and 5) efficient alert- ing mechanisms. Recent surveillance systems can be categorized into spe- cialized and generalized systems. The systems in [3–5] are specialized in train or railway surveillance [4], traffic or highway surveillance [3], and elevator surveillance [5]. These systems are tuned for specific application, therefore can not be deployed in different situations. Generalized systems can be further categorized into dis- tributed and non-distributed systems. Non-distributed or centralized systems deliver less performance due to the increased hardware requirements. On the contrary, dis- tributed systems [2, 6] utilize communication protocols to divide the work amongst a network of less powerful comput- ers, thus making automated surveillance systems available for residential or commercial use. To address the requirements of modern surveillance, this paper proposes a system with: 1) modular and extensible design, 2) distributed computing allowing the use of stan- dard computers, 3) MPEG-4 compression for reduced disk and network consumption, 4) real-time delivery of video over multicast, and 5) real-time event notification. The remainder of the paper is organized as follows: Sec- tion 2 gives an overview to the proposed system; Section 3 presents the proposed system in detail. Section 4 discusses experimental results and Section 5 concludes the paper. 2 System Overview The proposed system consists of three modules: Video Workstations, Control Workstations and a Server. Video Workstations are responsible for capturing the raw video from sources such as cameras, network streams or disks. Video Workstations are then responsible for processing the captured video to extract events from the scene. The pro- posed system uses the methods in [1] to extract surveillance events. The video signal is then compressed and sent us- ing multicast to Control Workstations via the Real Time Streaming Protocol (RTSP). A Video Workstation registers with the Server when it goes online. All communication, excluding surveillance video, between Video and Control Workstations is logged. The Server maintains security information about the access privileges of all connected Control Workstations and distin- guishes between different levels of access. The Server also transfers the recorded videos from the Video Workstations at a scheduled time. It also archives and stores the received video and provides the means to access them. Control Workstations, in a similar manner to the Video Workstations, register with the Server. A Control Work- station can view the feed from or give command to one or more Video Workstations. The proposed system uti- lizes embedded media players to view the surveillance video streams. Commands are created from the graphical user in- terface (GUI) events on Control Workstations and are sent to Video Workstations, through the Server. Control Work- stations are programmed to respond to events using alert mechanisms. In the proposed system, visual and text mes- saging alerts are used as an example. 3 Proposed System The proposed system consists of a set of modules (see Fig. 1): Video Workstations VW = {v i },i ∈{1, 2, ..., N VW }, where N VW is the number of Video Workstations, Con- trol Workstations CW = {c j },j ∈{1, 2, ..., N CW }, where N CW is the number of Control Workstations and a cen- tralized Server S. Each module is built from a set of com- ponents, which fall under two main categories. The first, generic components, denoted by ˆ C are reusable compo- 1-4244-0038-4 2006 IEEE CCECE/CCGEI, Ottawa, May 2006 1001