Coordinate fixes boost LiDAR pedestrian detection across real and synthetic data
Coordinate Frame Recovery, Per Class NMS and Null Return Modelling for Sim2Real V2X LiDAR Detection
Computer Vision and Pattern RecognitionRobotics
Summary
Detecting objects like pedestrians using LiDAR sensors trained only on computer-generated data can be tricky when used in real life. The authors found that fixing mismatches in how locations are reported, adjusting how overlapping detections are filtered per object type, and modeling missing sensor data significantly improved detection accuracy. Their work focused on improving these practical problems rather than building a new detection method. This led to better alignment between simulated training and real-world testing for roadside LiDAR.
What this means in practice
- •For autonomous vehicle developers: Improve pedestrian detection in roadside LiDAR systems trained on synthetic data by correcting coordinate frames and class-specific suppression settings.
- •For smart city infrastructure teams: Enhance accuracy of real-world roadside object detection sensors by simulating realistic sensor artifacts such as null data returns.
Authors
Aman Upganlawar
Abstract
We describe our approach to the V2X-Real track of the UrbanTwin Sim2Real LiDAR Challenge, in which a 3D object detector must be trained exclusively on synthetic data and evaluated on unseen real roadside LiDAR from an Ouster OS1-64 unit, while the synthetic point clouds are simultaneously scored for distributional realism. Our final entry scores 0.7457 combined, from a detection mAP of 0.4382 and a realism score of 0.9251. Three findings account for most of our result, and all three were diagnostic rather than architectural. First, a coordinate frame mismatch between our predictions and the evaluation ground truth was suppressing detection AP entirely; recovering it took our score from 0.3520 to 0.5147 in a single submission. Second, a single global non maximum suppression threshold was destroying genuine pedestrian detections, because pedestrians occur in groups and a 0.3 IoU threshold applied to a 0.7 m box suppresses distinct neighbouring people; correcting this per class raised overall mAP from 0.0751 to 0.4295. Third, approximately 31 percent of the real reference point cloud consists of null returns encoded as the literal point (0, 0, 0), a format feature entirely absent from our synthetic clouds, which pinned Earth Mover's Distance near its floor until we modelled it. Negative results are reported at equal length, and Section 9 gives a full account of data provenance and compliance. That section also accounts for every mechanism contributing to the submitted score.