Improving the Accuracy of Gait Detection Using Computer Vision

Abstract

Using computer vision for gait analysis is easier and more cost-effective to implement in the field than using wearable devices or motion capture, etc. OpenPose is one of the freely available skeletal detection algorithms, but the accuracy of skeletal detection is not always high. Therefore, in this paper, the gait cycle is derived from the skeletal coordinate data obtained by general OpenPose. Based on the gait cycle, we propose a method to predict the coordinates and correct the skeletal coordinates.