ALTHighway: Comprehensive ULS-Based Point Cloud Dataset for Road Condition and Surrounding Environment Analysis

This paper presents the ALTHighway dataset, an expansive collection of point cloud data derived from the G219 national highway in Xinjiang's Altay Prefecture, China. Our dataset contains approximately 17 million points acquired with unmanned LiDAR scanning and annotated in detail into five categories: road, slope, grass, vegetation and other. We evaluated the PointN et,
$\text{PointNet}++$
, DeLA, and its enhanced model on this dataset. The improved DeLA model demonstrated superior performance, achieving a mIoU of 73.0%, outperforming the previous DeLA model by 4.0 absolute percentage points. We hope ALTHighway will improve highway environment monitoring and analysis by providing extensive training data for 3D point cloud deep learning understanding tasks.