Planar segmentation from depth images using gradient of depth feature

In this work, a new algorithm for segmenting depth images into planar regions is presented. Developing a new algorithm for planar segmentation has been motivated by the results from other state of art methods which are tailored to segment depth images generated from Time of Flight cameras or structure light cameras but failed to segment depth images generated from stereo camera of normal textured objects. Unlike other state of Art methods which mainly use local surface normals for planar segmentation; a new Gradient of Depth (GoD) feature is proposed. The proposed GoD feature is parameter-free, easy to compute in terms of complexity and faster to compute compared to the local surface normals. The proposed GoD feature is implemented in a robust algorithm which uses the advantage of the 1D dimension feature space of the GoD feature and meets the performance of state of art algorithms in terms of quality segmentation. Moreover the proposed algorithm is able to segment planar regions from nonplanar objects and it is robust to parameter changes which allows to be used on different scenes generated using different cameras.