Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A Fully Interpretable Statistical Approach for Roadside LiDAR Background Subtraction

Aitor Iglesias, Nerea Aranjuelo, Patricia Javierre, Ainhoa Menendez, Ignacio Arganda-Carreras, Marcos Nieto

TL;DR

The paper tackles background subtraction for roadside LiDAR by introducing a fully interpretable statistical framework that models background via a Gaussian Distribution Grid learned from background-only scans. The approach combines GDG construction from accumulated and voxelized point clouds with a transparent, rule-based foreground/background classification pipeline, achieving sensor-agnostic performance across rotating and MEMS LiDARs. Evaluations on the public RCooper dataset show superior accuracy and robustness compared with state-of-the-art methods, even with limited background data, and demonstrate practical viability on low-resource hardware such as the Jetson Nano 2GB. The work emphasizes interpretability, scalability, and generalization, while outlining future directions toward real-time optimization and semantic foreground classification for infrastructure-assisted autonomous driving.

Abstract

We present a fully interpretable and flexible statistical method for background subtraction in roadside LiDAR data, aimed at enhancing infrastructure-based perception in automated driving. Our approach introduces both a Gaussian distribution grid (GDG), which models the spatial statistics of the background using background-only scans, and a filtering algorithm that uses this representation to classify LiDAR points as foreground or background. The method supports diverse LiDAR types, including multiline 360 degree and micro-electro-mechanical systems (MEMS) sensors, and adapts to various configurations. Evaluated on the publicly available RCooper dataset, it outperforms state-of-the-art techniques in accuracy and flexibility, even with minimal background data. Its efficient implementation ensures reliable performance on low-resource hardware, enabling scalable real-world deployment.

A Fully Interpretable Statistical Approach for Roadside LiDAR Background Subtraction

TL;DR

The paper tackles background subtraction for roadside LiDAR by introducing a fully interpretable statistical framework that models background via a Gaussian Distribution Grid learned from background-only scans. The approach combines GDG construction from accumulated and voxelized point clouds with a transparent, rule-based foreground/background classification pipeline, achieving sensor-agnostic performance across rotating and MEMS LiDARs. Evaluations on the public RCooper dataset show superior accuracy and robustness compared with state-of-the-art methods, even with limited background data, and demonstrate practical viability on low-resource hardware such as the Jetson Nano 2GB. The work emphasizes interpretability, scalability, and generalization, while outlining future directions toward real-time optimization and semantic foreground classification for infrastructure-assisted autonomous driving.

Abstract

We present a fully interpretable and flexible statistical method for background subtraction in roadside LiDAR data, aimed at enhancing infrastructure-based perception in automated driving. Our approach introduces both a Gaussian distribution grid (GDG), which models the spatial statistics of the background using background-only scans, and a filtering algorithm that uses this representation to classify LiDAR points as foreground or background. The method supports diverse LiDAR types, including multiline 360 degree and micro-electro-mechanical systems (MEMS) sensors, and adapts to various configurations. Evaluated on the publicly available RCooper dataset, it outperforms state-of-the-art techniques in accuracy and flexibility, even with minimal background data. Its efficient implementation ensures reliable performance on low-resource hardware, enabling scalable real-world deployment.
Paper Structure (14 sections, 1 equation, 2 figures, 5 tables, 1 algorithm)

This paper contains 14 sections, 1 equation, 2 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. State of the art
  3. Roadside LiDAR datasets
  4. Perception for roadside LiDAR
  5. Methodology
  6. Gaussian Distribution Grid generation
  7. Background subtraction
  8. Experiments
  9. Dataset
  10. Evaluation
  11. Comparative Evaluation
  12. Per-Class Evaluation
  13. Time Performance
  14. Conclusions and Future Work

Figures (2)

  • Figure 1: Diagram of our roadside LiDAR background subtraction algorithm. From top to bottom: (1) Gaussian Distribution Grid generation, which creates a statistical representation of the background using Gaussian distributions; this representation serves as the foundation for the subsequent background subtraction process. (2) Background Subtraction, which uses the learned Gaussian distributions to classify each LiDAR point as either background or foreground.
  • Figure 2: Qualitative results on the RCooper dataset RCooper with different sensor setups. (a) Corridor with a single multiline 360° LiDAR. (b) Intersection with a single MEMS LiDAR. (c) Intersection with multiple LiDARs ($2×$ 360° $+$$2×$ MEMS). Blue: background (TP), green: foreground (TP), red: background (FN), purple: foreground (FN). Top view (left) and oblique view (right).