Direct 3D pattern matching

Abstract

We have studied the problem of extracting shape parameters from freeform features in full 3D. The freeform features are typically encountered in laser-scanned data from physical parts or in point clouds from any source. Matching involves 1) the search for optimal position and orientation of the template shape and 2) the adjustment of a number (d) of intrinsic shape parameters. The type and total number of parameters (6+d), and hence the family of template shapes is determined by the type of feature. We have analyzed the robustness, accuracy and the efficiency of Hausdorff-like shape distance measures. A number of search strategies have been tested and were evaluated against convergence and computational performance. The practical relevance of the technique is addressed as well.