Quality-Aware Video Analytics in Edge Computing Environments

Mohammad Nazmus Sadat, Erwin Vargas-Alfonso, Rui Dai

Abstract: Edge computing has opened new doors for real-time video analytics applications due to its ability to offer significantly faster response times by processing videos near the source. However, the limited computing capabilities of edge devices can affect the quality of video analysis. While the existing literature has focused on minimizing latency, the quality aspect needs to be rigorously investigated to satisfy the high accuracy requirements of video analytics applications. This paper proposes a new quality-aware video analytics framework for edge networks. First, a real-time video analytics platform based on edge computing is designed and implemented. Then, the impacts of different factors on the quality and the overall latency of edge computing-based video analysis are investigated. Next, the trade-off between the video analysis quality and the latency, i.e., computation benefits at the edge vs. remote server, is studied. Finally, a quality-aware edge computing-based video analytics framework (QVAF) is designed to minimize the total latency while guaranteeing a required detection accuracy. Evaluation results show that QVAF could significantly improve response times across different detection accuracy requirements.