Dense scene 3D reconstruction using color based sampling with fusion of image and sparse laser

This paper investigates dense scene 3D reconstruction by fusing camera images and sparse laser data. This paper proposes color based sampling to improve discontinuity between objects, distance accuracy and smoothing of the same object. For robustness to light and camera noise, we apply mean shift filtering to camera image. Distance value is sampled from sparse laser data using color similarity. Kernel based cost function is suggested to estimate distance value from sampled element. We suggest iterative refinement module to find optimal depth data. Color based sampling algorithm is robust to laser noise caused by laser scattering at object edges. Results are presented to demonstrate our proposed algorithm which is robust to image and laser data noise.