Comparison of LiDAR Data Filtration Algorithms for Enhanced Situational Awareness Under Noisy Conditions

In adverse weather conditions like fog, rain, smoke, and snow, LiDAR sensor data can become corrupted by noise, leading to missed detections or false positives. This paper presents a comparison of multiple state-of-the-art LiDAR data filtration techniques, namely Radius Outlier Removal (ROR), Dynamic Radius Outlier Removal (DROR), and Low Intensity Outlier Removal (LIOR), by evaluating their performance in addressing simulated noise, as well as realistic noise introduced by fog, smoke, and rain. The comparison is conducted by analyzing well-defined performance metrics, including, accuracy, error, precision, recall, and F-Score. Our filtration results indicate that overall performance of DROR is superior to both ROR and LIOR filtration algorithms, however, in specific short-range LiDAR imaging scenarios, the performance of LIOR can be comparable with DROR. Furthermore, this study presents a novel endeavor by establishing the relationship between performance metrics and minimum number of neighboring points (K_min) for both ROR and DROR filtration techniques, utilizing different densities of simulated noise, as well as realistic noise introduced by foggy and smoky conditions.