Dense surface reconstruction based on the fusion of monocular vision and three-dimensional flash light detection and ranging

Abstract

A dense surface reconstruction approach based on the fusion of monocular vision and three-dimensional (3-D) flash light detection and ranging (LIDAR) is proposed. The texture and geometry information can be obtained simultaneously and quickly for stationary or moving targets with the proposed method. Primarily, our 2-D/3-D fusion imaging system including cameras calibration and an intensity-range image registration algorithm is designed. Subsequently, the adaptive block intensity-range Markov random field (MRF) with optimizing weights is presented to improve the sparse range data from 3-D flash LIDAR. Then the energy function is minimized quickly by conjugate gradient algorithm for each neighborhood system instead of the whole MRF. Finally, the experiments with standard depth datasets and real 2-D/3-D images demonstrate the validity and capability of the proposed scheme.

Depth Map Enhancement based on Improved Adaptive Total Generalized Variation for ToF Camera

“…It is a convenient strategy to overcome the drawbacks of ToF camera by exploiting the advantages of each of the camera. The markov random field (MRF) model and simple linear iterative clustering are employed to generate high-quality depth map by optimizing global energy function [12]. Park et al, [13] combine the MRF and the nonlocal means filtering to upsample the depth maps.…”