Abstract
Vehicle localization using roadside LiDARs can provide centimeter-level accuracy for cloud-controlled vehicles while simultaneously serving multiple vehicles, enhancing safety and efficiency. While most existing studies rely on repetitive scanning LiDARs, non-repetitive scanning LiDAR offers advantages such as eliminating blind zones and being more cost-effective. However, its application in roadside perception and localization remains limited. To address this, we present a dataset for infrastructure-based vehicle localization, with data collected from both repetitive and non-repetitive scanning LiDARs, in order to benchmark the performance of different LiDAR scanning patterns. The dataset contains 5,445 frames of point clouds across eight vehicle trajectory sequences, with diverse trajectory types. Our experiments establish baselines for infrastructure-based vehicle localization and compare the performance of these methods using both non-repetitive and repetitive scanning LiDARs. This work offers valuable insights for selecting the most suitable LiDAR scanning pattern for infrastructure-based vehicle localization. Our dataset is a significant contribution to the scientific community, supporting advancements in infrastructure-based perception and vehicle localization. The dataset and source code are publicly available at: https://github.com/sjtu-cyberc3/BenchRNR.
1. Introduction
Roadside vehicle localization is a cornerstone of Intelligent Transportation Systems (ITS), enabling centimeter-level accuracy for cloud-controlled vehicles. Most existing research utilizes repetitive scanning LiDARs, which feature rotating mirrors to create fixed horizontal scan lines. While effective, these sensors are costly and have inherent blind zones. Non-repetitive scanning LiDARs, such as the Livox Avia, use unique scanning patterns to cover the field of view more comprehensively over time, potentially eliminating blind zones at a fraction of the cost. However, the resulting point clouds are often sparse in a single frame, necessitating a benchmark to evaluate their performance in real-world infrastructure settings.
2. The Bench-RNR Dataset
We introduce Bench-RNR, a first-of-its-kind dataset specifically designed to benchmark these two LiDAR technologies. The data was collected from a building-mounted sensor suite overlooking a parking lot.
- Scale: 5,445 frames across 8 vehicle trajectory sequences.
- Sensors: A high-end Hesai 128-line repetitive LiDAR and two Livox Avia non-repetitive units.
- Ground Truth: 1-cm accuracy provided by vehicle-mounted high-precision GNSS/IMU.
- Diversity: Covers 11 different parking spots and various trajectory types.
3. Methodology
We implemented four localization baselines to compare performance:
- Seg+obb: PCA-based orientation estimation.
- Seg+convex: Convex hull fitting for bounding box estimation.
- PV-RCNN: A deep learning-based point-cloud detector pre-trained on DAIR-V2X-I.
- Register-Loc: A template-matching approach that aligns live point clouds with a high-resolution 3D vehicle model.
4. Results and Discussion
The primary challenge for non-repetitive LiDAR is point sparsity; a single frame yields significantly fewer points on target vehicles compared to high-line repetitive scanners. However, our experiments show that template matching (Register-Loc) effectively mitigates this.
Quantitative analysis reveals that non-repetitive LiDAR can achieve a mean localization error of 6.87 cm, nearly identical to the 6.84 cm error of the 128-line repetitive scanner. This performance is achieved despite a 100x difference in sensor cost ($1,400 vs $140,000), highlighting the potential for cost-effective smart city infrastructure.
5. Conclusion
Bench-RNR demonstrates that non-repetitive LiDAR is a viable, cost-efficient alternative for high-precision roadside localization. Future work will expand the dataset to include higher vehicle speeds and more complex urban environments.
Dataset and Code: https://github.com/sjtu-cyberc3/BenchRNR
Summary updated