Modular Approach for Online Vertical Obstacle Detection

Abstract

Poles, towers, and other vertical structures are a significant hazard to low-altitude flight operations. Three-dimensional sensors have the ability to perceive these objects; however, detection is hindered by the overwhelming preponderance of returns from other surfaces. This paper first evaluates existing approaches to finding vertical structures. Then the paper proposes a fast, modular process that efficiently removes extraneous points and automatically distills remaining data to find significant vertical structures with application to dense and sparse datasets. We apply our approach to the analysis of 200 million LiDAR points from a variety of real-world scenes. With our algorithm, the average density of identified pole returns for communication towers increases from 0.1% to over 60%. Finding other vertical structures is more difficult, but prevalence generally increases for these objects also.