An Improved Indoor Map Construction Method Based on Millimeter-Wave Radar

Aimed at enhancing accuracy and enriching semantic features of indoor map construction which is used for path planning of robot, this paper proposes a novel method based on extracting features from point clouds which come from millimeter-wave radar, combined convolutional neural networks. Data obtained from millimeter-wave radar are divided into two parts. One part extracts line segments and edge points as features of point clouds. The features build similar index is used to represent similarity between two point clouds to choose optimal transform matrix which transforms one point cloud into another. We take advantage of influences of different materials on millimeter-wave, using convolutional neural networks to distinguish different materials in environments to construct semantic maps. The results of experiments show that our method can achieve more accurate map than the typical map construction algorithm. The results also show that the obtained map can effectively represent materials categories of indoor environment.

I. Introduction

In recent years, with the rapid development of indoor robots, its concerning areas such map construction, navigation and path planning, are attracting increasing research attention. Especially, map construction plays a significant role as the global positioning system cannot be used in indoor environment. Under this situation, robots need to create maps by themselves so as to perceive environments and navigate autonomously.