A Study of Features for Action-Recognition Approach in Human Body with Occlusion

Abstract

Recently, Japan’s worker is keeping aging. Specifically, in construction site, it is necessary to have a safety-management system for older workers. Moreover, the system that identifies and collects information of behavior and physical condition from workers, including but not limited to older workers, will contribute to improving the efficiency and safety of work site. Currently, our research group discussed an action-recognition approach using skeleton data as a key part of the system, which for the purpose of developing work-management system that can recognize the action from workers. However, a situation usually happened that when workers moving and carrying in work size, some parts of body are shielded or blurred by objects. It became difficult to extract whole skeleton data and generate low action-recognition accuracy due to the losing of body parts. In this paper, we discussed the feasibility of features for action-recognition approach in human body with occlusion situation.