Year: 2025
Type: conference-paper
Abstract: This paper proposes a multi-modal fusion cooperative perception system that integrates LiDAR and camera data to enhance the environmental awareness of autonomous vehicles. The system overcomes the limitations of single-sensor methods, such as the lack of detailed visual information in LiDAR and the inability of cameras to accurately perceive depth. By combining the strengths of both sensors, the proposed fusion strategies significantly improve detection accuracy and robustness, especially in complex traffic scenarios. The paper introduces fusion mechanisms such as channel-level concatenation, element-level summation, and attention-based multi-head self-attention, which enhance feature fusion. Experiments on the OPV2V dataset demonstrate the effectiveness of the system, particularly in heterogeneous modality scenarios. The results show that the proposed system outperforms single-modality systems in collaborative perception tasks between multiple vehicles, achieving better detection accuracy and robustness. Finally, the paper highlights the potential of this approach for real-world autonomous driving applications and suggests further optimization of fusion strategies and exploration of additional sensor modalities for cooperative perception.
Authors: Qian Zuo, Chenzihan Li, Yulong Fan, Wenshan Kang, Tao Yang, Nana Feng, Zhen Tian, Mingyu Yu
Language: English
Subfield: Automotive Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W4412623693