Improved Object Matching Using Structural Relations

This paper presents a method for object matching that uses local graphs called keygraphs instead of simple keypoints. A novel method to compare keygraphs was proposed in order to exploit their local structural information, producing better local matches. This speeds up an object matching pipeline, particularly using RANSAC, because each keygraph match contains enough information to produce a pose hypothesis, significantly reducing the number of local matches required for object matching and pose estimation. The experimental results show that a higher accuracy was achieved with this approach.