눈이 오는 환경에서 라이다 외란을 제거하기 위한 반사 강도 기반 노이즈 제거 알고리즘 (LiDAR Noise Filtering in Snowy Environments Using Reflectance Intensity)

서지정보

· 발행기관 : 국방로봇학회
· 수록지 정보 : 국방로봇학회 / 4권 / 2호 / 20 ~ 27페이지
· 저자명 : 김형석, 류철민, 이민영, 박지혁

초록

Lidar sensors are widely used in autonomous driving, drones, and smart factories for their capability to quicklydetect objects at long distances and produce high-resolution images. However, their performance is significantlyhindered by noisy environments, such as during snow and rain. To mitigate this issue, research has exploredsensor fusion with radar and optical cameras, as well as distance-based filtering algorithms to clean Lidar pointclouds. Despite these efforts, achieving real-time, noise-free point cloud data remains a challenge, oftencompromising accuracy. Recently, there have been various attempts to improve both speed and accuracy. Notably,studies utilizing the intensity values from Lidar sensors, as opposed to traditional distance-based methods, haveshown promising results. Nonetheless, these methods face a speed reduction problem because they cannot adjustthreshold values dynamically in response to varying snow intensity. Therefore, this study proposes a fast andaccurate noise removal filter by predicting the threshold through regression based on fluctuating snow intensityvalues. When evaluated by distinguishing between snowy and non-snowy areas, the proposed model demonstratedsignificant improvements in both speed and accuracy over existing methods

영어초록

Lidar sensors are widely used in autonomous driving, drones, and smart factories for their capability to quicklydetect objects at long distances and produce high-resolution images. However, their performance is significantlyhindered by noisy environments, such as during snow and rain. To mitigate this issue, research has exploredsensor fusion with radar and optical cameras, as well as distance-based filtering algorithms to clean Lidar pointclouds. Despite these efforts, achieving real-time, noise-free point cloud data remains a challenge, oftencompromising accuracy. Recently, there have been various attempts to improve both speed and accuracy. Notably,studies utilizing the intensity values from Lidar sensors, as opposed to traditional distance-based methods, haveshown promising results. Nonetheless, these methods face a speed reduction problem because they cannot adjustthreshold values dynamically in response to varying snow intensity. Therefore, this study proposes a fast andaccurate noise removal filter by predicting the threshold through regression based on fluctuating snow intensityvalues. When evaluated by distinguishing between snowy and non-snowy areas, the proposed model demonstratedsignificant improvements in both speed and accuracy over existing methods

참고자료

· 없음