Reliable wall estimation method with RGB-D camera for wall following

In this paper, we figure out the problems of existing wall estimation methods using a 2D sensor and propose a reliable wall estimation method using a RGB-D camera. First, a point cloud is obtained by the RGB-D camera, and a surface normal is calculated at each point. Next, planes such as floor, right and left planes are classified by angle thresholding of the normal vector, where the thresholding angle is determined as an angle between the normal vector and the gravity. In this case, the right and left planes can be used as reference walls for wall following. Since the relatively long wall segment can be obtained by using just 1 frame of the depth images, the wall can be obtained reliably and rapidly.