Comparative usage of performance measures for object detection and tracking in video surveillance

초록

Performance evaluation has become an indispensable part of video surveillance systems, but it is a big challenge for researchers to decide which performance measure is more suitable for a given situation. In this paper, frame-based, object-based, and area-based evaluation paradigms are reviewed for quantifying different aspects of object detection and tracking in video surveillance systems. We also developed an annotation tool to generate Ground Truth XML and a video analytical performance evaluation tool (INHA-VAPE) to make the evaluation more user-friendly. The measures are implemented in the INHA-VAPE tool and can be easily chosen according the detailed usages described.