Autonomous Wheelchair Following Traffic Guard’s Instruction Based on Action Recognition

Abstract

With the development of autonomous driving technology, small vehicles such as autonomous wheelchairs are gradually becoming common. In this work, we propose an RGB camera-based action recognition system that allows autonomous wheelchairs to understand instructions from traffic guards. Our system locate traffic guards by using YOLO to identify traffic batons, and then utilize OpenPose to obtain human body 2D key points in real time to classify his/her action.