Conversion of depth images into planar laserscans considering obstacle height for collision free 2D robot navigation

Abstract

Mobile robots have become popular in many application areas over the last few decades. In order to perceive their environment in which they move autonomously, various sensors such as depth cameras are used. The processing of 3D information from depth images is very computationally expensive due to the large amount of data. However, for many mobile robots, navigation in a simplified 2D world is sufficient. For this purpose, the depth images of the environment can first be reduced to 2D information in the form of a laserscan line and then processed with algorithms for localization and mapping. This paper improves an existing algorithm that converts depth images into laserscans and tests it on a real robot in a real world scenario. In contrast to the original algorithm, the improved algorithm considers 3D information such as the height of the robot and obstacles when creating the laserscan line. This allows a mobile robot to navigate in a simplified 2D world without colliding with obstacles in the real 3D world. Since processing the 3D information is computationally expensive, the algorithm was optimized to be executable on low-cost single-board computers.