To recognize objects, one must have an internal representation of an object suitable for matching its features to image descriptions. There are several approaches to modeling, and this research uses a surface-based object representation distinguished by its use of shape segmentable surfaces organized in a subcomponent hierarchy. This approach provides the information needed to efficiently recognize and locate complicated articulated structures (like a robot) when using the input data described in Chapter 3.
This chapter also briefly describes the SMS modeling approach [70],
whose development was based on experience with the IMAGINE I system.