Deep ViDAR：CNN based 360°panoramic video system for outdoor robot visual navigation and SLAM

Abstract

Normally we use a laser radar to measure the area which the robot can move, while there is relatively fixed semantic information in outdoor environments, especially roads. With deep learning in a wide range of applications of semantic image segmentation, we believe that the image information panoramic video stream semantic segmentation, enabling the robot to navigate rely on cameras in most scenes. Our proposed system can be performed for segmenting the image information in each camera, combined with the panoramic image stitching, which can reduce the cost of the hardware of robot navigation.