A General Approach of Calculating Semantic Object Distances from Complex Point Clouds

RGBD sensors are widely used in current robot research as they have the capacities of measuring geometrical distances and providing dense point clouds. However, it is still a difficult problem to extract semantic object distances from these point clouds. To solve this problem, a novel approach based on BP neural network is proposed in this paper. As a start point, we first analyze the complexity of obtaining semantic object distances. Then, the principle of the proposed approach is introduced in detail. Besides, in this work, Tensorflow is used to implement the BP neural network based approach. To make the training/evaluation dataset, a special software tool is developed. Finally, an object with complex geometry has been selected as a test object, and the experimental results have verified the effectiveness of the approach.