Vision based sidewalk recognition for walking assistive robot in outdoor environments

Robot assisting elderly people to walk in outdoor environments is important especially in aging societies. Robot outdoor navigation is really challenging due to the large variety of environments and situations. For safe robot navigation, the robots must recognize the sidewalks while avoiding pedestrians and obstacles. In this paper, we propose a deep learning method for sidewalk recognition. We trained two Deep Networks algorithms Yolo and YOLACT and compare the performance for sidewalk recognition. The trained Deep Neural Networks are implemented to control the walking assistive robot developed in our laboratory navigate in outdoor environments. The real robot implementation of trained Deep Networks shows a good performance.