Terrain normalized scan context improves forest lidar place recognition
TNVD-SC: Diagnosing Scan Context Failures in Forests with Terrain-Normalized Vertical Distributions
Computer Vision and Pattern RecognitionRobotics
Summary
Recognizing locations using 3D laser scans (LiDAR) in forests is hard because trees look very similar and the ground isn’t flat. The authors improve a common method called Scan Context by adjusting how vertical information is measured relative to the terrain. This new way better distinguishes places by encoding how tree heights vary with the ground, leading to more accurate matching of locations. Their improvements reduce storage and computation needs without losing accuracy.
What this means in practice
- •For forest mapping teams: Improve location recognition accuracy for long-term LiDAR scans in forested environments using terrain-normalized scans.
- •For robotics developers: Enhance robot localization performance in uneven, tree-covered areas by integrating terrain-relative vertical LiDAR descriptors.
Authors
Jineon Kim, Jhonghyun An
Abstract
Long-term LiDAR place recognition in forests is difficult because repetitive vegetation and uneven terrain create severe structural aliasing. We diagnose Scan Context (SC) by separating candidate retrieval from fine matching: on the standardized TreeLoc++ Wild-Places protocol, a TNVD support key places a valid location in the top-25 set for approximately 99% of queries, while conventional SC matching on those candidates reaches only 61.84% mean R@1. With an identical raw-SC candidate set, terrain normalization improves max-height cells by only 1.21 points but vertical-distribution cells by 19.72 points. These observations motivate Terrain-Normalized Vertical Distribution Scan Context (TNVD-SC), which preserves the SC ring-sector topology while encoding terrain-relative vertical probability distributions. Under the standardized TreeLoc++ protocol, TNVD-SC reaches 91.83% mean R@1 and attains the highest R@1 mean and R@1 stability among the compared methods, while TreeLoc++ remains stronger in mean MR, MF1, and AUC. The unchanged method reaches 85.14% mean R@1 on the four-session CVTNet-style NCLT evaluation. A lossless count-coded implementation with exact hierarchical Jensen-Shannon search reduces descriptor payload by 3.86x and compute latency by 23.8% without changing the audited retrieval outputs.