Algoritmo para la detección de zonas poco texturizadas en imágenes reales, para su uso en sistemas de visión estereoscópica

Abstract

The use of stereoscopic vision systems in no-structured real scenes is a common task today. One feature in real images is the presence of low-textured regions with no enough information to select a correct match between two pixels. Local algorithms reduce remarkably his performance in this kind of images. In this work we show an algorithm for detection of low-textured zones, this algorithm avoids to find incorrect matches, the incorrect matches are marked for a posterior process. Results show an improvement in the accuracy on the disparity maps when this algorithm is used.