Dense Depth Map Estimation Based on Camera–LiDAR Sensor Fusion

This paper proposes a camera-LiDAR sensor fusion-based dense depth map estimation method for robust and accurate depth estimation in real-world driving environments. The proposed method consists of five stages: pre-processing, weighted clustering, surface modeling, point cloud sampling, and semantic segmentation-based image processing. Surface modeling and point cloud sampling enhance the density and representation of objects in sparse point clouds, while a segmentation mask effectively preserves object boundaries. Experimental results demonstrate that the proposed method achieves superior performance compared to existing methods in terms of MAE. In particular, cumulative distance-wise MAE analysis confirms that our method achieves higher accuracy in medium and long-range regions. Moreover, the proposed method consistently achieves stable performance across various vehicle detection models, demonstrating its robustness and applicability. Finally, we collected sensor data from diverse real-world driving environments, including road, city, and campus scenarios, to construct a dense depth map dataset and perform 3D reconstruction, demonstrating through qualitative results that the proposed method consistently reconstructs objects without noticeable distortions.