Pedestrian Positioning in Surveillance Video using Anthropometric Properties for Effective Communication

Positioning of pedestrians or persons is an important technique for video-based systems. For network surveillance systems, positioning can be applied to reduce data volume for storage devices and communication traffic. In this paper, we propose a simple positioning method using anthropometric properties such as a face length. A foot point in an image is estimated based on face detection results and anthropometric properties, then perspective transformation converts the foot point into a position on the floor plane. We improve the anthropometric model to reduce estimation errors of positioning. Moreover, as an application of pedestrian positioning, we implement data reduction functions of video data for surveillance systems. Experiments using a 4K video indicate that the average positioning error improves to 0.5 m. In terms of data reduction, we found that combination of tracking, selection of key frames, cropping, resizing, and JPEG compression reduce the 35.6 MB video data to 70 kB. These experiments induce that our approach realize simple and precise positioning and data reduction for effective communication for video surveillance systems.