ITSC 2025 Paper Abstract

Paper TH-LA-T17.2

Li, Luxi (Beijing Normal University - Hong Kong Baptist University United ), Li, Yuchen (Beijing Normal University - Hong Kong Baptist University United ), Ai, Yunfeng (University of Chinese Academy of Sciences), Tian, Bin (Chinese Academy of Sciences Institute of Automation), Xuanyuan, Zhe (Beijing Normal University-Hong Kong Baptist University United In), Chen, Long (Chinese Academy of Sciences)

SegMine: A Multi-Modal Segmentation Dataset and Benchmarks for Visual Understanding in Open-Pit Mines

Scheduled for presentation during the Invited Session "S17c-Synthetic-Data-Aided Safety-Critical Scenario Understanding in ITS" (TH-LA-T17), Thursday, November 20, 2025, 16:20−16:40, Southport 2

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Keywords Deep Learning for Scene Understanding and Semantic Segmentation in Autonomous Vehicles, Advanced Sensor Fusion for Robust Autonomous Vehicle Perception, Sensor Integration and Calibration for Accurate Localization in Dynamic Road Conditions

Abstract

Autonomous driving has made significant progress in recent years, yet its application in unstructured environments like mining remains underexplored. Mining areas, characterized by rugged terrain, indistinct road boundaries, and frequent accidents, present a critical unstructured scenario where autonomous systems could enhance operational safety. However, development is hindered by the severe shortage of mining-specific datasets and fundamental differences from urban scenarios, particularly the absence of explicit navigational cues like structured roads,lane markings or traffic signs. To address these issues, this paper introduces SegMine, the first publicly available mining-specific segmentation dataset comprising annotated images and point clouds. We propose a task-driven labeling paradigm to extract implicit navigational information unique to mining contexts, and benchmarks on image and lidar segmentation are provided. Additionally, we design Confusion-Weighted IoU (CW-IoU), a novel evaluation metric that improves task-aware assessment. Our work could advance scene understanding in unstructured environments and demonstrate practical potential for autonomous mining operations. The dataset is available at: https://github.com/DeSCI-aca/SegMine.