Data-driven child behavior prediction system based on posture database for fall accident prevention in a daily living space

Ten thousand children are admitted to emergency rooms due to accidents every year in Tokyo. The most frequent accident is a fall accident. Fall accidents may occur when climbing to a high place in a daily living space. Since injury prevention by human supervision does not work well, the World Health Organization recommends an environmental modification approach as an effective preventive countermeasure to this problem. Predicting children’s behavior is necessary in order to improve the environment. However, even for advanced human modeling technology, predicting where children can climb in everyday life situations remains difficult. In the present study, the authors developed a new method for predicting places that children can climb in a data-driven manner by integrating cameras, a behavior recognition system (OpenPose), and a climbing motion planning algorithm based on a rapidly exploring random tree. Thirty five children participated in an experiment to collect climbing posture data. A simulation is performed based on the posture database and allows us to visually understand how children climb up in daily living space. This makes it possible to improve to achieve a safe environment for children without the need for specialized knowledge, which is useful for parents, nursery teachers, nurses, etc. The present paper describes fundamental functions of the developed system and presents an evaluation of the feasibility of the prediction function.