Technical gazette, Vol. 32 No. 5, 2025.
Original scientific paper
https://doi.org/10.17559/TV-20240730001888
MSMFAM-VoxelNeXt: LiDAR-Camera Fusion for Highway Traffic Perception
Chunsheng Zhang
; Guangdong Provincial Highway Construction Co., Ltd, Guangzhou, Guangdong Province, China
*
Bibo Liu
; Boshen Branch of Guangdong Boda Expressway Co., Ltd, Guangzhou, Guangdong Province, China
Changwei Wang
; Boshen Branch of Guangdong Boda Expressway Co., Ltd, Guangzhou, Guangdong Province, China
* Corresponding author.
Abstract
Roadside perception is critical for intelligent transportation systems, but faces challenges in sensor fusion and data processing. This paper proposes an enhanced perception scheme integrating LiDAR and camera data. We introduce a multi-scale multi-feature attention module (MSMFAM) to enrich voxel features, addressing issues of voxel size and semantic information extraction. Point cloud levelling and data simulation augmentation techniques improve detection accuracy across varying sensor heights. Our fusion algorithm combines LiDAR and image results with elliptical matching for enhanced target detection and classification. Experimental results show significant improvements over baseline algorithms, with mAP increases of 2.2% in point cloud detection and 1.5% infusion results. The proposed method demonstrates potential for advancing roadside perception in intelligent transportation systems.
Keywords
fusion perception; intelligent transportation; point cloud detection; roadside perception
Hrčak ID:
335056
URI
Publication date:
30.8.2025.
Visits: 256 *