Our paper titled "Uirapuru: Timely Video Analytics for High-Resolution Steerable Cameras on Edge Devices" has been accepted to the 31st Annual International Conference On Mobile Computing And Networking. MobiCom is the top conference in mobile computing and networking and it will be held in Hong Kong, China this year.
The paper addresses the challenge of real-time deep-learning-based video analytics on high-resolution cameras that support pan-tilt-zoom (PTZ) actuations. We propose Uirapuru, an efficient runtime system that supports adaptive tiling of video frames based on object distributions while incorporating a comprehensive understanding of the camera actuation. Our evaluation with both real-world and synthetic camera feeds shows that Uirapuru can achieve 1.45× improvement in accuracy while respecting specified latency budgets or reaches up to 4.53× inference speedup with on-par accuracy compared to state-of-the-art approaches.
Uirapuru: Timely Video Analytics for High-Resolution Steerable Cameras on Edge Devices
Guilherme Henrique Apostolo (Vrije Universiteit Amsterdam, The Netherlands), Pablo Bauszat (Vrije Universiteit Amsterdam, The Netherlands), Vinod Nigade (Vrije Universiteit Amsterdam, The Netherlands), Henri Bal (Vrije Universiteit Amsterdam, The Netherlands), Lin Wang (Paderborn University, Germany)
