The input to verification is a fully instantiated object hypothesis.

SURFACEs necessarily exist as they are inputs to recognition, so only ASSEMBLY existence needs to be verified. The goal is to reject hypotheses that are coincidental, which means showing that the surfaces associated with the hypothesis cannot be organized into a solid. Solidity is based on complete connection of all visible surfaces, which requires a topological examination of the evidence.

Identity is based on object-specific properties. The level of detail for most previous three dimensional object recognition systems was superficial and so an object meeting the criteria was identified as far as the computation was concerned, but, unfortunately, not for us as observers. Here, identification is complete to the level of description embodied in the model, so increasing the level of verification entails increasing the level and structure of the evidence. Hence, associated with each model is a set of constraints that the data must satisfy, and any structure that meets these is accepted as a model instance.

The main goal of verification is to reject false hypotheses, and this ultimately requires comparison between the object and model shapes. This can be done efficiently by comparing the symbolic characterizations of the surfaces (the boundaries, the curvature axes and the curvature magnitudes) and the relationships of these features in the object reference frame.

Following ACRONYM [42], it seems sensible to specify property requirements
using numerical constraints on the values of an object's properties.
Each satisfied property gives some measure of certainty and together they
help ensure correct identification.

- SURFACE Verification
- Rigid ASSEMBLY Verification
- Non-rigidly Connected Object Verification
- Numerical Constraint Evaluation