Object-Detection-Based Video Compression for Wireless Surveillance Systems
Lingchao Kong, Rui Dai · IEEE Multimedia · 2017
Many distributed wireless surveillance applications use compressed videos for automatic video analysis tasks. However, the accuracy of object detection--which is essential for video analysis--can be reduced because lossy compression degrades video quality. Current standardized video-encoding schemes can cause temporal fluctuation for encoded blocks in stable background areas of a raw video, which strongly affects object-detection accuracy. To obtain better object-detection performance on compressed videos, the authors introduce a standards-compliant video-encoding scheme that can suppress unnecessary temporal fluctuation in stable background areas. New mode-decision strategies, designed for both intra- and interframes, reduce the temporal fluctuation while maintaining acceptable rate-distortion performance. Experimental results show that, compared with traditional encoding schemes, the proposed scheme improves object-detection performance and results in lower bit rates with comparable quality.