TileClipper: Lightweight Selection of Regions of Interest from Videos for Traffic Surveillance

Explore a 15-minute conference talk from USENIX ATC '24 that introduces TileClipper, an innovative system for efficient traffic surveillance video processing. Learn how this approach utilizes tile sampling to significantly reduce bandwidth requirements while maintaining high detection accuracy. Discover how TileClipper adaptively selects rectangular areas within video frames based on their correlation with tile bitrates, resulting in an average 22% reduction in data sent to the cloud while achieving 92% detection accuracy. Gain insights into the system's performance on various datasets, its real-time capabilities on affordable edge devices like Raspberry Pi 4 and NVIDIA Jetson Nano, and its live deployment results showing over 87% detection accuracy and 55% bandwidth savings. Understand how TileClipper addresses the challenges of processing video feeds from thousands of traffic cameras without the need for expensive re-encoding techniques.

USENIX ATC '24 - TileClipper: Lightweight Selection of Regions of Interest from Videos...