Edge Detection on Indoor Scenes Using Local Binary Pattern Features

To detect the edge information and extract whole planes in the input indoor scene RGB-D images is an important and fundamental issue for scene analysis and understanding. This paper presents a novel edge detection method by using the local binary pattern features computed on scene deep images. First, the gradient information along X-direction and Y-direction are calculated in the scene deep map and the basic edge information can thus be detected using these two gradient maps. Then, the local binary pattern features nearby the basic edges can be computed and the normal information of the indoor scene point cloud can also be calculated. Finally, combining the local binary pattern features and the normal information, we can adjust and extract the edge information in the indoor scene. Experimental results illustrate that our proposed approach can effectively detect the edge information of indoor scenes, and also avoid the over-detection and under-detection during edge extraction.