Depth-map driven planar surfaces detection

Abstract

Planar surface is a common feature in man-made structure, thus accurate detection of planar surface can benefit the image/video segmentation and reconstruction and also the navigation system of robots. Since depth map represents the distance from one object to the capturing camera in a grey image, it also can represent the surface characteristics of the objects. So in this paper, we propose a novel Depth-map Driven Planar Surface Detection (DDPSD) method, where detection starts from "the most flat" seed patch on the depth map and uses dynamic threshold value and surface function to control the growing process. Compared with one of the popular planar surface detection algorithms, RANdom SAmples Consensus (RANSAC), the accuracy of the proposed method is obviously superior on typical indoor scenes. Moreover, semi-planar surfaces can be also successfully detected by the proposed method.