Modelagem Paralela em C+CUDA de Sistema Neural de Visão Estereoscópica

摘要

The images formed on our retinae are bidimensional; however, from them our brain is capable of synthesizing a 3D representation with color, shape and depthinformation about the objects in the surrounding environment. For that, after choosing a point in 3D space, our eyes verge to this point and, at the same time, the visual system is fed back with the eyes position information, interpreting it as the distance of this point to the observer. Depth perception around the vergence point is obtained using visual disparity, i.e., the difference between the positions in the retinae of thetwo projections of a given point in 3D space caused by the horizontal separation of the eyes. Most of the depth perception processing is done in the visual cortex, mainlyin the primary (V1) and medial temporal (MT) areas. In this work, we developed a parallel implementation in C+CUDA of model, built at UFES, of the neural architecture of the V1 and MT cortices that uses as building blocks previous modelsof cortical cells and log-polar mapping. A sequential implementation of this model can create tridimensional representations of the external world using stereoscopic imagepairs obtained from a pair of fronto-parallel cameras. Our C+CUDA parallel implementation is almost 60 times faster and allows real-time 3D reconstruction.