Development of Human and WheelChair Combined Pose Estimation Model with Machine Learning

Abstract

Recently, human pose estimation using machine learning has been used in various fields. In the field of sports, human pose estimation has been applied to motion analysis and skill evaluation in baseball, tennis, soccer and other sports. In wheelchair sports, it is possible to derive an index to evaluate wheelchair skills by combining the pose estimation data of a human and a wheelchair. However, there are no reports of pose estimation including wheelchairs. The purpose of this study is to develop a pose estimation model that combines human and wheelchair from a RGB camera image. Based on the human pose estimation model, we retrain the model using a dataset that includes feature points of a human and a wheelchair, and evaluate the model. Furthermore, we proposed a method for detecting wheelchair push motion from pose estimation data, and evaluated it using a 3-axial accelerometer and video.