Accuracy Evaluation of Human Detection in Different Resolutions of Surveillance Videos

Abstract

We aim to provide a real-time, high-quality and low network load video surveillance system. In this paper, for an early stage, we evaluate an accuracy of human detection in different resolutions of surveillance videos by using state-of-the-art human detection algorithms: YOLOv2 and OpenPose. Evaluation results conclude that adaptive control of video resolution may improve precision value or recall value.