We present a vision system to recognize planar objects composed of lines on condition of partial view and large image distortion by perspective effect. Under projective transformation, we cannot use euclidean geometric features such as length, angle and size ratio of lines. As a result, the recognition problem becomes a very difficult task.
Our system uses the projective invariants of five lines on a plane to find a corresponding model stored in a database. However, a set of five coplanar lines produces 120 different invariant values depending on the order of five lines. Therefore, the search space to find a corresponding model in the database becomes so huge when many models are considered.
We introduce the concept of line convex hull defined by five lines. Line convex hull gives a unique ordering to a set of five lines under projective transformation, so they have a single invariant value. In addition,this classifies them into 21 geometric types.
ILF(Indexing Logic Filter) indexes models using line convex hull. ILF produces indexing logic value and order as index keys. They have a very strong power in discrimination of data. Therefore search space can be reduced remarkably.
Also we present a recognition method of 3D objects, which are composed of planar panel and stiffeners. To do this, we select planar lines in 3D lines of an object using a stereo camera system. And then we recognized 3D objects in 2D recognition system by the selected planar lines.
In series of real experiments using fifteen prototype panels, the system successfully recognizes panels in the database.
