One-Shot Deep Learning-Based Camera–LiDAR Extrinsic Calibration with Pyramid Dilation-Based Encoder

The integration of cameras and light detection and ranging (LiDAR) in multisensor systems has made precise extrinsic calibration increasingly critical. Although deep learning methods have demonstrated potential, they typically struggle to fuse RGB images with LiDAR depth data effectively in dynamic or complex environments, particularly for models with high parameter counts and computational overhead. We propose a framework that overcomes these challenges by dynamically leveraging multimodal data with relevant and fewer parameters to deliver real time, robust, and scalable calibration. The proposed framework fully exploits the complementary strengths of both sensors while ensuring low resource consumption and fast inference, making it ideally suited for deployment in resource-constrained autonomous systems.