Robust Outlier Removal from Point Clouds Acquired with Structured Light

Abstract

We present a method for automatic, robust identification of outliers in point clouds acquired with structured light. In contrast to most state of the art methods, we consider available 2D information instead of operating on the points only. Our method performs robust even for complex, glossy surfaces, where illumination artifacts introduce ghost geometry or outliers that seamlessly blend into the correct surface. The key idea of our algorithm is to only use cameras for reconstruction and the projector for a separate consistency check. We use encoding by phase shifting to obtain per pixel correspondences for the cameras and subpixel accuracy for the projector. The method requires at least 2 cameras, a projector and a precise calibration of all devices.