Implementation of Dynamic Radius Outlier Removal (DROR) Algorithm on LiDAR Point Cloud Data with Arbitrary White Noise Addition

Abstract

In autonomous vehicles or robots, the LiDAR sensor plays the most efficient role in achieving the best environmental perception due to its superior resolution and field of view. Under adverse weather conditions, such as fog, mist, rain, or snow, undesired sparse measurement points blend into the original point cloud, which in turn make an effect in the forms of missing detections or false positives. Under the heaviest precipitations, these undesired sparse measurement points, also known as noise points, can even be misinterpreted as an object by an autonomous vehicle or robot, which brings the autonomous entity to a complete stop. Surprisingly, very little research has been carried out to filter out the noises in the LiDAR point cloud data under adverse weather conditions. In this paper, two of the most efficient LiDAR data filtration techniques reported, i.e., Radius Outlier Removal (ROR) and Dynamic Radius Outlier Removal (DROR), have been compared using an existing LiDAR point cloud. For the very first time, under different concentrations of added noise, the effectiveness of DROR filtration algorithm on the LiDAR point cloud has been demonstrated. For each of the concentrations of noise, DROR filtration yields an effective filtration method based on performance metrics analysis and by keeping the original LiDAR data intact and removing most of the noise points. Additionally, a large number of noise points, which pertain to an extreme weather condition, has been generated, followed by the implementation of DROR filtration on a highly noisy LiDAR point cloud. DROR filtration reduces very few points on the original LiDAR point cloud and almost every noise point has been removed over a good range of LiDAR (nearly 70 meters), which indicates an effective filtration in extreme weather condition.