Probabilistic depth map fusion for real-time multi-view stereo

In this paper we propose a probabilistic method for fusing depth maps in real time for wide-baseline situation. We treat the depth map fusion as a problem of probability density function (pdf) estimation. The original point cloud, instead of the reprojected depth map, is used to estimate the pdf, and a mathematical expectation computation method is proposed to reduce the complexity of the method. Experimental results show that the proposed method can get the fused depth map in real time, and is very promising for fusing depth maps from multiple depth cameras with sparsely distributed viewpoints.