Abstract
This paper presents a new categorization method for typical indoor places (e.g., "office", "kitchen", "corridor") using 3D scans by a laser range finder. The proposed technique utilizes panoramic range and reflectance images obtained by a laser range finder. Firstly, the panoramic range and reflectance images are divided into several partial images, and histograms of local binary patterns (LBP) are created from each partial image. The place category of each partial image is determined by support vector machines using a feature vector which concatenates the histograms. Finally, the place category for a panoramic image is determined by voting the decision of partial images. Experimental results show that the recognition rate of the place categorization is improved by the proposed method using panoramic images compared with the method using partial images.
