Streaming quality control based on object-detection accuracy

This study aims to establish an adaptive video quality control method that satisfies object-detection accuracy and reduces bandwidth consumption simultaneously for remote real-time video analysis systems. Existing video quality control methods determine video quality by considering human perceptual characteristics; this reduces bandwidth consumption while providing a high quality of experience. However, it cannot reduce bandwidth consumption in systems that use object-detection engines to detect people and vehicles. Thus, this study proposes a video quality control method to reduce bandwidth consumption. To this end, the bandwidth consumption and ratio of the number of frames satisfying the mean average precision requirements to the total number of frames (herein referred to as the success rate) are evaluated. The results confirm that the proposed method can reduce bandwidth consumption to 49% of that of the existing video quality control method at the same success rate.