Three-dimensional laser radar navigation method and equipment in glass scene based on deep learning

Abstract

The invention discloses a three-dimensional laser radar navigation method and equipment in a glass scene based on deep learning. The navigation method comprises the following steps: performing offlinetraining to obtain a glass recognition deep neural network and an optical characteristic deep neural network; in the navigation process, inputting the glass probability recognized by the glass recognition deep neural network and the calculated glass direction into a space voxel map in real time; in a space voxel map containing the glass probability and the glass normal vector, simulating a laserbeam emitted by the laser radar by adopting a light projection method, obtaining the probability that the laser beam transmits the glass based on an optical characteristic deep neural network to screen radar points, and generating a simulated point cloud set; and finally, based on the real three-dimensional radar data and the simulation point cloud set, calculating the absolute pose of a robot inthe space voxel map by adopting normal distribution transformation. According to the navigation method, perception confusion based on laser radar positioning can be avoided, and the three-dimensionalnavigation precision of the mobile robot in a glass scene is improved.

Description

technical field [0001] The invention belongs to the field of mobile robot navigation, and relates to a deep learning-based three-dimensional laser radar navigation method and equipment in a glass scene, and more specifically, to a deep neural network-based mobile robot three-dimensional laser radar navigation method in a glass scene. Background technique [0002] LiDAR-based navigation is one of the most reliable and accurate navigation methods to obtain robot pose. Recently, glass walls have gradually become the main elements for mobile robots navigating buildings such as museums or shopping malls. However, unlike opaque objects, glass has reflective properties that can lead to blurry or erroneous lidar perception processes. Therefore, it is difficult to apply lidar-based localization to scenes with glass walls. In order to improve the positioning accuracy of the robot under various environmental conditions, it is necessary to develop a reliable lidar-based positioning me...

AI Technical Summary

Problems solved by technology

But normal distribution transforms have difficulty matching lidar data to maps in glass environments due to perceptual confusion caused by glass

However, existing localization systems cannot efficiently obtain glass information from voxel maps, which degrades localization performance.

Images