Principal > Alumnos > Octavio Israel Rentería Vidales

Octavio Israel Rentería Vidales

   »  Generación: Septiembre2017
   »  Grado: Maestría en Computación
   »  Asesor: Juan Carlos Cuevas Tello
   »  Co-asesor: Mariano José Juan Rivera Meraz
   »  Línea de Investigación: Ninguna
E-mail | Webpage | Currículum

Convolutional Neural Networks on an embedded system applied to optical flow


The optical flow is the measurement of the displacements of common pixels between two images, it can be in time (video) or in stereo images (left and right), this last type of displacement can be used to create 3D reconstructions that can be useful to detect obstacles in autonomous vehicles. Traditional methods for flow estimation are based on differential or discrete optimization methods, but recently new algorithms based on Convolutional Neural Networks (CNN) such as FlowNet have appeared, based on a encoder-decoder architecture, have prove that a CNN is a viable option for this problem. The CNN implementation for low-consumption hardware is also an important issue, with the advent of concepts such as IoT (Internet of things), the need for this light systems to be able to make decisions in place arises. The purpose of this research is to create a CNN for optical flow estimation in stereo images. This network should be light enough to work on low power hardware and with enough precision that it can be used on small autonomous robots.