Automated event recognition for video surveillance systems
LE3 .A278 2003
Master of Science
Video surveillance systems need capabilities that go beyond simple recording and retrieving by date and time. Automated event recognition and scene understanding are becoming increasingly important to deal with the explosion of surveillance system data and the shortage of security personnel to analyze it. This thesis provides an overview of automated event recognition approaches for single stationary video camera surveillance systems. The motion detection module analyzes raw video and provides moving "blobs". The object tracking module tracks moving object, gives each object a unique identification and provides object properties. The event recognition module detects all kinds of events based on object properties and relationships among objects. Atomic events are defined to represent states of objects at an instant in time. Most atomic events can be detected directly from object properties. Several atomic events compose a basic event that represents change of state of an object during a very short period of time, usually less than one second. Complex events describe scenarios that usually happen over a longer period of time. A complex event is composed of several sub-events that may be basic events or other complex events. A deterministic finite state machine (DFA) is chosen to recognize complex events by matching object behavior with predefined complex event models. Besides video surveillance, the approaches described in this thesis can also be used to broader areas, such as scene understanding and image analysis.
The author grants permission to the University Librarian at Acadia University to reproduce, loan or distribute copies of my thesis in microform, paper or electronic formats on a non-profit basis. The author retains the copyright of the thesis.