A simple information fusion method provides the obstacle with saliency labeling as a landmark in robotic mapping

Abstract

Mapping is the key technology of the robot slam algorithm, mainly used to obtain environmental geometric information. How to effectively improve the ability to obtain semantic information on the map becomes essential. Salient object segmentation with depth can be used as a kind of salient semantic information that helps the robot better understand the environment. This paper put forward a consistent salient determination method based on visual and lidar images for complementary advantages. Our method is based on the simple information fusion of the two sensors' salient object key features, including approximate centroids and contour widths. Theoretical analysis and experimental results demonstrate that the proposed method has the advantage of strong illumination robustness and can work under weak light conditions, which can solve the task in the visual image when the front salient object blocks the back salient object. It can be used to detect salient objects with approximate size in indoor and outdoor sparse scenes. The detected salient objects can be used as semantic labeling labeled in the environment and as the relative position reference for helping the robot improve the accuracy of mapping, positioning, and planning the motion more intelligently.

Keywords

• Salient semantic information
• Consistent salient determination
• Information fusion
• Semantic labeling