Multi-Sem Fusion: Multimodal Semantic Fusion for 3-D Object Detection

LIDAR and camera fusion techniques are promising for achieving 3-D object detection in autonomous driving (AD). Most multimodal 3-D object detection frameworks integrate semantic knowledge from 2-D images into 3-D LiDAR point clouds to enhance detection accuracy. Nevertheless, the restricted resolution of 2-D feature maps impedes accurate reprojection and often induces a pronounced boundary-blurring effect, which is primarily attributed to erroneous semantic segmentation. To address these limitations, we present the multi-sem fusion (MSF) framework, a versatile multimodal fusion approach that employs 2-D/3-D semantic segmentation methods to generate parsing results for both modalities. Subsequently, the 2-D semantic information undergoes reprojection into 3-D point clouds utilizing calibration parameters. To tackle misalignment challenges between the 2-D and 3-D parsing results, we introduce an adaptive attention-based fusion (AAF) module to fuse them by learning an adaptive fusion score. Then, the point cloud with the fused semantic label is sent to the following 3-D object detectors. Furthermore, we propose a deep feature fusion (DFF) module to aggregate deep features at different levels to boost the final detection performance. The effectiveness of the framework has been verified on two public large-scale 3-D object detection benchmarks by comparing them with different baselines. And the experimental results show that the proposed fusion strategies can significantly improve the detection performance compared to the methods using only point clouds and the methods using only 2-D semantic information. Moreover, our approach seamlessly integrates as a plug-in within any detection framework.