Multiview scene matching using local features and invariant geometric constraints

Abstract

A novel scene recognition method that utilizes local appearance descriptions together with geometrical invariants for multiview scene matching is presented in this paper. The rationale behind this effort is to complement the lowered discriminative capacity of local features, with invariant geometric descriptions. Presented method is evaluated by comparison with a prominent baseline method, which utilizes Random Sample Consensus (RANSAC) for robust 2D affine invariant transform estimation. Experimental results have revealed the superiority of the presented method which utilizes 3D geometric invariants over the baseline robust 2D transform estimation method, especially in typical scenes for which planarity assumption does not hold.