Um método escalável para recuperação de objetos 3D utilizando árvore de vocabulário

Abstract

Interpreting a scene from the computational perspective, usually is summarized by the interpretation and classification of its elements, realizing how they are organized by the scenario. This is done by recognizing the models presented as the input to a system that looks for similar models in a database. This process is named models retrieval, where matching and indexing are usually part of the process. Although technology has developed, the robust interpretation of scenes from the environment is still a challenge for computers in general, hindering areas that use Computer Vision, such as Robotics. Given the need for new shapes retrieval algorithms, this work aims to deepen the studies in the area of Computer Vision, proposing a new algorithm capable of recovering threedimensional shapes in large databases. Although the similarity criteria are important, the indexing of characteristics for the search of similar objects also have their importance and directly impacts the speed and quality of the problem. For this reason, the focus of this work is to propose a new method for the retrieval of three-dimensional models, using the data structure called vocabulary tree, which presents significant results in the literature for queries by two-dimensional models. In this work, the method is presented in detail and is implemented through an algorithm. The results obtained by this algorithm are presented and analyzed using databases and metrics, which were used in previous works. The method presents promising results, but with large room for improvement.