Content area
This dissertation considers the problem of optimal video compression for video surveillance of vehicle traffic. In modern video surveillance systems the specific context of traffic surveillance is rarely considered, which results in systems that require large bandwidth channels or perform poorly in automated object tracking applications.
We develop a low complexity optimization framework that automatically identifies video features critical to tracking and concentrates bitrate on these features. We develop methods to realize such optimization via Region of Interest (ROI) coding, non-uniform transform coefficient quantization tables, pre- and post-processing, and algorithms combining these methods. In order to quantify our results, and in order to perform iterative optimization, we develop a unified tracking accuracy metric. Using the H.264/AVC video coding standard and commonly used state-of-the-art trackers we show that our algorithms allow for up to 90% in bitrate savings while maintaining comparable tracking accuracy.
We show the performance of our system in the presence of packet losses due to simulated wireless channel fading. We propose an algorithm using bitrate savings from our algorithms and H.264/AVC Redundant Slices to minimize this effect while maintaining most of the bitrate savings.
Keywords: Video Compression, Transportation Surveillance, Object Tracking, Transform Coding, Video Segmentation, Video Pre-Processing, Video Post-Processing