Effect of Video Transcoding Parameters on Visual Object Tracking for Surveillance Systems

In any video surveillance system, it is very important to provide effective remote viewing to heterogeneous users with various network conditions and viewing device. To meet this adaptability requirement, the video transcoding process is inevitable, which consists of converting a video from one compressed format to another. However, since the transcoding operation is a lossy process, this can effect the performance of video analysis techniques such as visual object tracking. Consequently, in this paper, we evaluate the impact of video transcoding parameters on the performance of visual object tracking algorithms. To address this, first, a large-scale transcoding surveillance video (TSV) dataset is constructed. Then, we design a framework for assessing the performance of trackers. Finally, we evaluate six state-of-the-art trackers on the TSV dataset and analysis their performance with regard of transcoding parameters. Experimental results show that the video transcoding can negatively affect the performance of visual object tracking.