Research and Application of Semantic Point Cloud on Indoor Robots

The obstacle avoidance method based on visual and ultrasonic sensors and corresponding algorithms are widely researched and applied among route planning of indoor robot. This paper proposes a semantic point cloud construction method based on FC-DenseNets to analyze and compare the filtering performance of different parameters selection which being involved in the Statistical Outlier Removal filter. The original point cloud data is obtained by the RGB-D camera, and the best parameters setting as K=20 and α=1.0 is determined as conclusion. To achieve better image segmentation, FC-DenseNets and deep learning method is combined to contribute the point cloud with semantic information, which providing more richer environmental information for the robot’s mobile obstacle avoidance. Experimental results show that the proposed algorithm outperforms on simplified structure and distinct segmentation of environmental information.