Design and Implementation of an Optical Flow-based Autonomous Video Surveillance System

Autonomous video surveillance systems typically consist of several functional modules working in concert. These modules perform specialized tasks including motion detection, separation of the foreground and background, depth estimation, object tracking, feature estimation, and behavioral analysis. Computational overhead and redundancy may result from designing each module individually, as each module may incorporate different variety of techniques and algorithms. This paper presents the design of a surveillance system that uses an optical flow algorithm throughout. We consider the capabilities, solutions, and limitations of this design. Additionally, an evaluation of the performance of optical flow in specific situations, such as depth estimation, rigid and non-rigid classification, segmentation, and tracking, is provided. The main contribution of this work is a new system-level architecture based on a single key algorithm (optical flow) for the entire video surveillance system.