Shape-Based Recognition of 3D Objects in 2D Images

An object-recognition algorithm analyzes data from two-dimensional (2D) images to locate and identify possibly complexly shaped three-dimensional (3D) objects in possibly highly cluttered scenes depicted in the images. More specifically, the algorithm implements a relatively simple, effective, and fast process for recognizing 2D objects that may be partly occluded and that have shapes that can be modeled by use of sets of line segments (see figure). Because the algorithm tolerates a fair amount of perspective distortion, it is also applicable to 3D objects represented by sets of viewpoint- dependent 2D models.