Hardware aspects of a real-time surveillance system

In English This paper presents the implementation of an automated digital video surveillance system on an FPGA platformfor real-time performance. To achieve real-time performance, the system includes hardware accelerators for video segmentation, morphology operations, labeling and feature extraction while tracking is handled in software. By implementing a complete system on a hardware platform, bottlenecks in computational complexity and memory requirements can be identified and addressed. A memory access reduction scheme for the video segmentation part that utilizes pixel locality is proposed which shows the potential of reducing accesses with > 60%. Furthermore, a low complexity morphology architecture with low memory requirements is presented together with a labeling unit based on a contour tracing technique. Some thoughts on which features that can be extracted throughout the system are discussed.