Constructing High-Precision Geometric Models from Sensed Position Data

The construction of geometric scene models from sensed data is a fundamental task in image understanding. The creation of such models is essential in computer vision systems for applications ranging from reconnaissance to robot navigation. Automated model construction using visual techniques can also provide important support for systems in manufacturing and simulation. Vision-based geometric modeling has typically emphasized generality, with little explicit concern for the accuracy of the models produced. When high-precision is required, domain-specific representations and processing can provide significant advantages. We describe such a system for reverse engineering mechanical parts as part of maintenance and repair activities. Many of the key components of this system are also applicable to the generation of models for simulation and virtual environment applications.