FABWF-LGAR: Fully adaptive Bi-modal weighted fusion with lightweight gated attention refinement for efficient camera-LiDAR 2D object detection

Advanced Driver Assistance Systems (ADAS) engage multi-modal sensor data, including cameras and LiDAR, to recognize their environment accurately. However, fusing various sensor data while maintaining computational efficiency is yet major research. This work introduces Feature-Adaptive Bidirectional Weighted Fusion with Lightweight Gated Attention Refinement (FABWF-LGAR), a lightweight framework created to proceed multimodal sensor fusion for autonomous driving. The proposed work syndicates feature from different sensors with a bidirectional weighted fusion method that differs the contribution from each type of data. A lightweight gated attention enhancement module is fused to focus on suitable spatial and semantic features while reducing noise and redundant information. The proposed method increases the quality of the fused data without raising the computational complexity. The framework directs to enhance perception tasks in ADAS, like object detection and scene recognition, while being suitable for real-time use on resource-limited vehicles. The proposed FABWF-LGAR achieves 55.93% mAP@0.5 on the KITTI validation set, outperforming camera-only and LiDAR-only baselines by 34.5 and 50.7 percentage points, respectively. Experimental results show that this fusion policy specifies stronger and more efficient feature integration compared to traditional methods, highlighting its potential for real-world autonomous driving systems.