A Visual-attention-based 3D Mapping Method for Mobile Robots

Abstract

Human visual attention is highly selective. The artificial vision system that imitates this mechanism increases the efficiency, intelligence, and robustness of mobile robots in environment modeling. This paper presents a 3-D modeling method based on visual attention for mobile robots. This method uses the distance-potential gradient as motion contrast and combines the visual features extracted from the scene with a mean shift segment algorithm to detect conspicuous objects in the surrounding environment. This method takes the saliency of objects as priori information, uses Bayes' theorem to fuse sensor modeling and grid priori modeling, and uses the projection method to create and update the 3-D environment modeling. The results of the experiments and performance evaluation illustrate the capabilities of our approach in generating accurate 3-D maps.