Análise da geração de mapas de disparidade estéreo a partir de um algoritmo baseado em otimização

Abstract

The stereo correspondence problem has been studied for many researchers in computer vision. This research proposes to evaluate an optimization algorithm for obtaining disparity maps in terms of quality and processing time and to find parameters that can improve them. For that, the optimization algorithm was implemented in C++ using the opencv library. The algorithm was tested using the Middlebury dataset, which provides ground truth for evaluation comparison. Moreover, the processing time was also evaluated. As a result, it was found that at a certain time of execution, the error stabilized, and it is not necessary to continue executing iterations. In this research, it was verified that two parameters of the proposed algorithm, the SAD maxime mask size and the disparity average mask size, have influence in the quality of the disparity maps results. In terms of processing time, there was not such important variations, but an influence by the SAD maxime mask size for the tested scenarios.