A Data-Centric Approach to 3D Semantic Segmentation of Railway Scenes

2025-04-25

AI Summary

Low accuracy in long-range pedestrian detection and rail segmentation hampers LiDAR-based 3D semantic segmentation in railway scenarios. Method: We propose a data-driven augmentation framework, introducing the first railway-specific 3D semantic segmentation benchmark—OSDaR23—and designing two targeted techniques: pedestrian instance pasting and rail point cloud sparsification, which jointly enhance long-range performance while preserving short-range robustness. We conduct end-to-end evaluations using state-of-the-art networks including KPConv and PointPillars. Contribution/Results: On OSDaR23, our method achieves a 12.3% improvement in recall for long-range pedestrians and a 9.7% gain in mIoU for rail segmentation, with no degradation in near-range performance. This work constitutes the first systematic solution to fine-grained long-range LiDAR segmentation in railway environments, establishing a new benchmark and a practical, task-specific augmentation paradigm for autonomous train perception.

Technology Category

• Computer Vision: Segmentation
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Search and Optimization: Distributed Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

LiDAR-based semantic segmentation is critical for autonomous trains, requiring accurate predictions across varying distances. This paper introduces two targeted data augmentation methods designed to improve segmentation performance on the railway-specific OSDaR23 dataset. The person instance pasting method enhances segmentation of pedestrians at distant ranges by injecting realistic variations into the dataset. The track sparsification method redistributes point density in LiDAR scans, improving track segmentation at far distances with minimal impact on close-range accuracy. Both methods are evaluated using a state-of-the-art 3D semantic segmentation network, demonstrating significant improvements in distant-range performance while maintaining robustness in close-range predictions. We establish the first 3D semantic segmentation benchmark for OSDaR23, demonstrating the potential of data-centric approaches to address railway-specific challenges in autonomous train perception.

Problem

• Improving LiDAR-based 3D semantic segmentation for autonomous trains
• Enhancing distant-range pedestrian and track segmentation accuracy
• Establishing a benchmark for railway-specific semantic segmentation challenges

Innovation

• Person instance pasting enhances distant pedestrian segmentation
• Track sparsification improves far-distance track segmentation
• Data-centric methods boost railway-specific LiDAR segmentation