Reconhecimento eficiente de objetos usando multifoveamento em nuvem de pontos 3D

Abstract

Technological innovations in the field of hardware and 3D sensors allowed real time 3D point clouds acquisition. Therefore, varieties of interactive applications related to the 3D world that have been receiving increasing attention from researchers, arisen. However, one of the main problems that remains is the computationally intensive processing that requires optimized approaches to deal with this 3D vision model, especially when it is necessary to perform tasks in real time. Thus, we started from a proposed 3D multiresolution model presented as foveated point clouds which is a possible solution to this problem, but is limited to a single foveated structure with context dependent mobility. In this way, our proposal is an improvement of this model with the incorporation of multiple foveated structures. However, the union of several foveated structures results in a considerable increase of processing, since there are intersections between regions of distinct structures, which are processed multiple times. We address this problem by using a proposed multifoveated model that regards intersections on the union procedure. Such approach can be used to identify objects in 3D point clouds, one of the key tasks for automation, with efficient synchronization, allowing the validation of the model and verification of its applicability in the context of computer vision. The results demonstrate a gain in performance of the proposed model in relation to the use of multiple structures of the foveated point cloud model.