Mobile robot indoor semantic mapping using 3D laser scanning and monocular vision

This paper provides a method for indoor semantic mapping in 3D environment. For indoor environment constructed by numerous planar surfaces, plane features are extracted and classified to build the main structure of indoor scene. To identify and cognize different objects located in indoor scene, both the position information and the color information are used in object classification. After the background structures of indoor scene such as walls and floor are eliminated, a clustering algorithm is used to accomplish spatial segmentation in 3D laser scanning data. According to the prior color histograms of different objects stored in the database, we can effectively identify those objects associated with new spatial segmentation by matching their histograms with each one in the database, so that mobile robot could implement indoor semantic map building robustly. Experiment results implemented in real mobile robot platform show the validity and practicability of the proposed method in this paper.