Mobile 3D visual search using the Helmert transformation of stereo features

This paper presents a scheme for mobile 3D visual search that facilitates mobile recognition of 3D objects. We use a multi-view approach to extract the 3D geometric information of the query objects and integrate it into SIFT descriptors. To meet a given transmission bandwidth, we use a rate-constrained quad-tree representation for feature selection and encoding. With this approach, we are able to progressively match the query features against the stereo features in the database and implement a robust geometric verification with the Helmert transformation. The experimental results show that our proposed approach to mobile 3D visual search outperforms JPEG and single-view SIFT-based search.