雷达学报

This paper proposes a novel multimodal collaborative perception framework to enhance the situational awareness of autonomous vehicles. First, a multimodal fusion baseline system is built that effectively integrates Light Detection and Ranging (LiDAR) point clouds and camera images. This system provides a comparable benchmark for subsequent research. Second, various well-known feature fusion strategies are investigated in the context of collaborative scenarios, including channel-wise concatenation, element-wise summation, and transformer-based methods. This study aims to seamlessly integrate intermediate representations from different sensor modalities, facilitating an exhaustive assessment of their effects on model performance. Extensive experiments were conducted on a large-scale open-source simulation dataset, i.e., OPV2V. The results showed that attention-based multimodal fusion outperforms alternative solutions, delivering more precise target localization during complex traffic scenarios, thereby enhancing the safety and reliability of autonomous driving systems.

来源：2024年第1期

《雷达学报》期刊编辑部

查看雷达学报杂志2024年第1期