Object Pose Recognition Using Triangulation-Weighted RANSAC

The patent performs preliminary triangulation on all matched feature point pairs before applying RANSAC. By constructing triangles from corresponding feature points in template and actual images and calculating their intersection points, the method pre-identifies likely correct matching pairs. This preliminary action filters out many wrong matches before RANSAC begins, allowing the algorithm to converge much faster even when wrong match ratio exceeds 50%.

The patent introduces triangulation and intersection point calculation as an intermediary step between feature point matching and RANSAC transformation matrix calculation. The intersection points of triangles formed by corresponding feature points serve as a mediator that provides geometric constraints to identify correct matches. This intermediary mechanism bridges the gap between raw feature matches and reliable pose estimation, enabling RANSAC to work efficiently even with high error rates.

The patent performs preliminary triangulation and intersection point calculation to pre-identify correct feature point pairs before RANSAC. By using geometric constraints from triangle intersections, the method pre-filters matching pairs to ensure higher quality input for RANSAC. This preliminary action maintains robustness against mismatching while significantly improving the likelihood of obtaining an accurate transformation matrix even when wrong match ratio exceeds 50%.

The patent replaces the purely statistical random sampling mechanism of traditional RANSAC with a geometric constraint-based selection method. Instead of relying solely on random sampling and iterative validation, the method uses triangulation geometry and intersection point analysis to deterministically identify correct feature point pairs. This substitution of geometric reasoning for random mechanical sampling improves both success rate and precision of transformation matrix calculation.