Gait Recognition with 2D Pose Infomation Using a Surveillance Camera

Preventive care management requires the evaluation of functional training and influencing factors to set clear goals. Gait information on frail elderly individuals is useful for evaluating the effectiveness of functional training; however, till date, no general method exists to monitor this gait information. Our objective is to develop a technology for monitoring gait information in the long term by automatically associating the person from camera images using person matching technology. In a previous study, we studied person matching using 2D pose estimation. However, this evaluation used a camera temporarily placed at the same height as the elderly. To enable long-term monitoring, the camera must be permanently installed. We develop a person matching method that can be used with images from a surveillance camera, imaging at a depression angle installed at a typical elderly facility. The proposed method achieves 93.6% precision, suggesting its applicability in practice.