Visual recognition of planar panels with line features

The authors introduce a new method for visual recognition of panels with stiffeners for the sub-assembly process of ship-building. Recognition with existing CAD data is to find the location/orientation of stiffeners and panels, panel ID numbers and approximate welding lines in sequence. The method extracts line features from original images and uses five-line invariants for the recognition. The existing invariant approaches have been proved successful for planar object recognition, but the large search space with combinatorial explosion has been the main hurdle in real application. To reduce this search space, the authors propose a new method with line convex hull (LCH) and indexing logic filter (ILF). In addition, a hierarchical database is constructed to retrieve corresponding model features directly. They have performed experiments using fifteen different welding panels, and demonstrate the feasibility of their method for the recognition of planar objects.