Confidence-Aided Disparity Map Refinement with LiDAR-Stereo Fusion

We propose a disparity map refinement technique that synergizes the benefits of stereo cameras and LiDAR sensors. Our method dynamically adjusts the contributions of each sensor based on the confidence map from the RGB image. The inclusion of an image reconstruction loss function aims to reduce the reliance on LiDAR, advancing a robust disparity map estimation. Our experiments, which involve manipulating the LiDAR distribution, showed the decreased dependency on LiDAR. We evaluated our approach using the public benchmark, where it demonstrated commendable performance. Notably, the method is both efficient and straightforward in its implementation, underscoring its potential in leveraging the combined strengths of stereo and LiDAR information to achieve precise disparity map estimation.