Person Matching Technology using Gait Information of 2D Pose Estimation

The proportion of the elderly in Japan continues to increase, and there is concern about a shortage of caregivers. Therefore, a mechanism is needed to extend the healthy life expectancy and reduce the nursing care period. Since falling is one of the major causes of bedridden life, we aim to provide a service to plan appropriate gait training for elderly people at high risk of falling. The system has developed that extracts gait information by OpenPose and determines fall risk in order to monitor long-term in a way with less burden on the elderly. In addition, for the automation of monitoring, a function to match the same person from video images of different dates and times and record gait information is necessary. From the viewpoint of cost, we would like to utilize two-dimensional gait information by OpenPose for collecting gait information. This paper reports the development of a method matching the same person in different videos by using two-dimensional gait information based on OpenPose to automate long-term monitoring. This paper also reports the evaluation of the conventional matching methods.