Expansion-Based Depth Map Estimation for Multi-View Stereo

This paper presents an algorithm for acquiring high-quality models from multiple calibrated photographs by computing and merging depth maps. The algorithm first computes depth maps from multi-view stereo using a proposed expansion-based approach that returns a 3D point cloud with noisy and redundant information. Then the estimated depth maps are merged into an accurate surface model by a cleaning, downsampling, surface normal estimation and Poisson surface reconstruction process. The proposed approach has been implemented and the experimental results with several real datasets demonstrate that the approach can produce accurate surface models efficiently.