Deep Learning based Detection, Segmentation and Vision based Pose Estimation of Staircase

Mobile robots can move around in the surrounding. They are used in military, industrial applications, for surveillance tasks, etc. These tasks involve multi-floor navigation through the staircase. For efficient and safe stair climbing, it is necessary for the robot to align itself with the staircase. This paper presents a deep learning based approach for staircase detection, semantic segmentation of stair edges and vision based techniques with statistical operations for pose estimation of the staircase. The objective is to solve the problem of staircase traversing for mobile robots without any prior knowledge of the stair geometry and location by iteratively calculating the pose of the staircase from the robot’s point of view till it gets aligned with the staircase. The experimental results of staircase detection, semantic segmentation and pose estimation algorithms are presented further in the paper.