Shape estimation from 3D point clouds

Abstract

The problem of estimating a shape in a 3D point cloud data is important due to its general applicability in image analysis, computer vision, and graphics. It is challenging because the data is typically noisy, cluttered, partly missing and unordered. We address shape estimation using a template object under a fully statistical model, where the data is assumed to be modeled using a Poisson process on the object’s boundary (surfaces), corrupted by additive noise and a clutter process. Using analytical likelihood function dictated by the model, we optimize over pose and scale associated with hypothesized templates and estimate most likely shapes in observed point clouds under given shape hypotheses. We demonstrate this framework using examples of 2D and 3D shape estimation in simulated and real data.