자율주행 차량 물체 식별 정확도를 위한 이웃 반사 강도 기반 라이다 점군 눈 입자 제거 필터

Abstract

Purpose: This study focuses on developing an algorithm that can sift through noisy LiDAR data in adverse weather and filter out snow points without losing essential details. By achieving this, we can boost the reliability of autonomous navigation systems in snowy conditions. Methods: We developed a novel filtering technique that considers the LiDAR intensity from surrounding points, not just the point of interest. We tested this method using the winter adverse driving dataset (WADS), applying our algorithm to LiDAR data distorted by snowy conditions. Results: This study determined the efficiency of our filter based on the degree of noise it removed and the number of essential points it preserved. The results demonstrated a significant improvement in data quality while keeping the most relevant information intact. Conclusion: The new filtering method offers a significant upgrade over previous studies on LiDAR, especially in maintaining crucial LiDAR data. This breakthrough paves the way for more dependable autonomous vehicle navigation in weather that typically disrupts sensor accuracy.