l Relationships from Two-Dimensional Drawings by Planning
Tsê Chêng, Holmes Hull · 1991
On interpretation a three-dimensiona l object by labelling edges of the object from its two-dimensional image is an important step in computer vision. A new method using planning to solve the line-drawing labelling problem is presented. The labelling problem can be formulated as the Constraint Satisfaction Problem (CSP) and has been proved to be NP complete. In this paper, a new labelling scheme, based on Huffman-Clowes's junction categories, is introduced to solve the problem using constrained resource planning methodologies called Four-Corner CRP model proposed by Keng& Yun. The experiments show that our approach can speed up labelling by reducing the number of backtracks in searching a consistent label between a pair of junctions because of planning in advance.