Dupopt: A Redundancy Video Analysis Mechanism
Wenwen Li, Yongxiang Zhao, Chunxi Li · 2021 7th International Conference on Computer and Communications (ICCC) · 2021
The video analysis based on Deep Neural Network (DNN) model has been widely deployed to provide various services in mobile applications. DNN-based video analysis with high accuracy requires massive computation power beyond the hardware capability of mobile devices. Offloading to an edge server or remote cloud server is a promising solution to solve the above challenge. However, offloading computation leads to higher network latency which will reduce analysis accuracy. This paper uses redundant transmission to reduce network latency in video analysis and proposes a mechanism named Dupopt, which can select the optimal combination of copy number and frame resolution to maximize analysis accuracy. Numerical results show that the mechanism Dupopt can significantly improve the accuracy of video analysis compared with the transmission mechanism without redundancy.