Beam-Wise Statistical Background Subtraction for Static Roadside LiDAR: A Cross-Sensor Benchmark Study

Alexander Baumann; Marcel Vosshans; Thao Dang

cs.CVAug 14, 2026

Beam-Wise Statistical Background Subtraction for Static Roadside LiDAR: A Cross-Sensor Benchmark Study

Authors: Alexander Baumann, Marcel Vosshans, Thao Dang

Organizations: Institute for Intelligent Systems which is part of the Faculty of Computer Science and Engineering, University of Applied Sciences Esslingen, Germany

Abstract

Background subtraction is a key preprocessing step for infrastructure-based LiDAR perception, enabling efficient isolation of dynamic traffic participants without semantic annotations. However, systematic cross-sensor evaluations and reproducible studies for static roadside LiDAR are missing. This paper presents a comparative benchmark of beam-wise statistical background subtraction for statically mounted LiDAR sensors. We formulate background estimation as a per-beam temporal modeling problem and investigate complementary statistical strategies that capture dominant as well as multi-modal background structures, combined with spatial filtering in the angular and 3D domain. To enable reproducible evaluation, we introduce HighwayScene, a new multi-LiDAR dataset recorded in a static roadside setup, and extend the public CoopScenes dataset with static/dynamic point-wise annotations. Across multiple scenes and heterogeneous sensing technologies, we demonstrate that beam-wise statistical modeling provides a robust and transferable solution. Combining lightweight per-beam models with spatial consistency filtering substantially improves precision while maintaining high recall and real-time capability. All datasets, annotations, and implementations are publicly released.

Figures & tables

Fig. 1: Example result of the density-aware coarse-fine triangle algorithm (CFTA) with spatial consistency filtering on the CoopScenes dataset. Correctly classified points are shown in light gray (background) and red (foreground), while misclassified points are shown in cyan.

Fig. 2: Overview of the HighwayScene dataset. The figure presents representative labeled frames from three statically mounted LiDAR sensors based on rotating, FMCW, and solid-state sensing principles (background: blue, foreground: red). Insets depict the corresponding scan patterns, illustrating differences in angular sampling and field-of-view. Combined with the use of ROI-based and velocity-based annotation strategies, these characteristics create heterogeneous conditions for evaluating background subtraction methods.

Fig. 3: Example of a beam-wise range histogram with two competing peaks: a dominant near-range peak caused by dynamic foreground observations and a smaller but farther peak corresponding to the static background. The original CFTA formulation selects the dominant peak, whereas the proposed two-stage peak analysis identifies the farther background peak and determines the correct background threshold.

Fig. 4: Example of a beam-wise range distribution modeled by a Gaussian mixture. Two components correspond to static background structures at different distances, while an additional component represents dynamic foreground observations. During inference, a measurement is classified as background if it lies within a gated distance

z_{\text{gate}}

of at least one component previously identified as background.

Sensor	Method	No Consistency Filtering			With Consistency Filtering
		P	R	F2	P	R	F1
Ouster	CFTA [ 5 ]	91.68	96.07	95.15	91.95	96.01	93.94
	Density-Aware CFTA	92.48	99.09	97.69	97.75	98.69	98.22
	Quantile-Constrained GM	91.39	99.54	97.79	99.59	98.92	99.25
	Beam-Wise GMM	94.00	99.74	98.53	99.47	99.39	99.43
	Image-Based GMM [ 18 ]	89.54	95.40	94.17	99.51	94.55	96.97

TABLE I: Quantitative performance of geometry-based background subtraction methods on the HighwayScene dataset across three LiDAR technologies. Results are reported without and with spatial consistency filtering. Background models are optimized for recall-weighted

F_{2}

, while filtering parameters are optimized for

F_{1}

. Precision (P), Recall (R), and corresponding

F_{2}

F_{1}

scores are given in %.

Consistency Filter	P	R	F1	RT
None	76.59	98.73	86.26	–
Radius (per-point)	98.56	96.64	97.59	5.95
Grid (range)	97.34	97.14	97.24	22.41
Grid (model)	94.03	92.70	93.36	38.74

TABLE II: Spatial consistency filtering strategies on HighwayScene (AEVA) using the beam-wise GMM. Precision (P), Recall (R), and

F_{1}

scores are reported in % and runtime (RT) in ms.

	Scenes
Method	Scene 1 (FG=3.14%)			Scene 2 (FG=0.78%)			Scene 3 (FG=0.59%)			Scene 4 (FG=1.53%)
	P	R	F1	P	R	F1	P	R	F1	P	R	F1
CFTA [ 5 ]	65.84	94.76	77.70	50.41	97.62	66.48	48.52	95.59	64.37	50.20	93.61	65.35
Density-Aware CFTA	94.84	92.81	93.82	92.69	93.55	93.12	92.62	90.42	91.51	99.61	93.78	96.61
Quantile-Constrained GM	96.72	88.41	92.38	91.73	91.10	91.42	95.35	82.46	88.44	98.76	91.35	94.91
Beam-Wise GMM	97.78	77.93	86.73	92.76	91.20	91.97	90.72	81.47	85.85	98.41	95.09	96.72

TABLE III: Background subtraction performance on four CoopScenes scenes with varying foreground ratios (FG). Results are reported per scene using Precision (P), Recall (R), and

F_{1}

score in %.

Robotic Perception

Robotic perception enables robots to interpret scenes, objects, motion, and contact from cameras and other sensors to guide action.

See all 320 papers

MR-LiDAR: A Multi-Resolution Roadside LiDAR Benchmark for Perception Diagnostics and Deployment Guidance
May 23, 2026Shunlai Cui, Peng Cao, Yuan Zhu +5Benchmark Design 3D Object Detection
School of Transportation and Logistics, Southwest Jiaotong University, Chengdu, 610031, China · Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA, 23529, USA · School of Transportation and Logistics, Southwest Jiaotong University +5
RESOLVE: A Multi-Resolution and Multi-Modal Dataset for Roadside Cooperative Perception
Jun 30, 2026Shaozu Ding, Linan Song, Marco De Vincenzi +1Autonomous Driving Datasets Camera-LiDAR Fusion
The Polytechnic School, Arizona State University, USA · Department of Computer Science and Engineering, New York University, USA
ATLAS: A Large-Scale Evaluation Benchmark for Adversarial LiDAR Perception
Jun 1, 2026Mellon M. Zhang, Siddhant Panse, Zimo Fan +3Lidar-Based 3D Object Detection Adversarial Attacks
Georgia Institute of Technology

May 23, 2026J/K move · Enter open · S save

MR-LiDAR: A Multi-Resolution Roadside LiDAR Benchmark for Perception Diagnostics and Deployment Guidance

Shunlai Cui, Peng Cao, Yuan Zhu, Yongjiang He+4

School of Transportation and Logistics, Southwest Jiaotong University, Chengdu, 610031, China · Department of Electrical and Computer Engineering, Old Dominion University, Norfolk, VA, 23529, USA · School of Transportation and Logistics, Southwest Jiaotong University+5

Benchmark Design 3D Object Detection

Open paper AskPDF arXiv

LiDAR configuration selection in roadside sensing remains constrained by limited controlled evidence across sensors. To address this gap, we present MR-LiDAR, a controlled cross-configuration benchmark comparing four LiDAR configurations from 16 to 128 beams on dynamic vehicle and vulnerable road user (VRU) targets. Its cross-configuration main set covers matched daytime, nighttime, and vehicle-occlusion conditions, while three deployment and operating condition subsets examine mounting height, target speed, and curved-road geometry. By reducing confounding from scene and trajectory variation, MR-LiDAR links sensor configuration to target point acquisition and algorithmic detection outcomes. We characterize distance-dependent point loss, define effective beam count as the number of vertical beams intersecting the target's angular extent, and introduce a point-based Occlusion Ratio. The results show that 16- and 32-beam configurations degrade earlier with distance, while VRUs exhibit greater point sparsity than motor vehicles. Nominal beam count alone is insufficient to explain roadside scanning capability. In the cross-configuration main set, the 80- and 128-beam configurations achieve comparable results, consistent with their nearly identical downward beam layouts. A uniform 80-beam counterfactual yields only about one third of the effective beam count of the actual non-uniform configuration. Illumination and the tested speed range have little influence on target-point-count patterns, whereas mounting height and near-field road geometry can substantially alter configuration ordering. Finally, an empirical minimal-sufficient configuration matrix combines frame-level geometric sufficiency with matched-box quality and includes threshold sensitivity analysis, providing bounded configuration guidance for the tested conditions.

Abstract

Robotic Perception

Explore similar work

MR-LiDAR: A Multi-Resolution Roadside LiDAR Benchmark for Perception Diagnostics and Deployment Guidance

RESOLVE: A Multi-Resolution and Multi-Modal Dataset for Roadside Cooperative Perception

ATLAS: A Large-Scale Evaluation Benchmark for Adversarial LiDAR Perception