Reliable 3D surface acquisition, registration and validation using statistical error models

We present a complete data acquisition and processing chain for the reliable inspection of industrial parts considering anisotropic noise. Data acquisition is performed with a stripe projection system that was modeled and calibrated using photogrammetric techniques. Covariance matrices are attached individually to points during 3D coordinate computation. Different datasets are registered using a new multi-view registration technique. In the validation step, the registered datasets are compared with the CAD model to verify that the measured part meets its specification. While previous methods have only considered the geometrical discrepancies between the sensed part and its CAD model, we also consider statistical information to decide whether the differences are significant.