Visión artificial aplicada a la detección e identificación de personas en tiempo real

Abstract

This degree work makes a study and a practical application of four different methods; Principal Component Analysis, Linear Discriminant Analysis, Elastic Bunch Graph Matching and Machine Learning; in image processing for real’s time faces detection and identification. The proposed methods for the study in this work were pre-selected in a planning stage, where it was made a brief actualization about the state of the art in order to know the best options for the goal’s accomplishment. Also, the proposed methods were implemented and tested, and among them, the solution with the best results was the Google’s convolutional neural network for CV called TensorFlow. Each method was tested according to the same conditions, but Google’s TensorFlow was the one that met the goals. Then, the selected method was also subjected to a new test where it was possible to determine the quantity of images required in order to reach at least the 90% in rates of detection, precision and accuracy. Tests were taken considering 4 different image capture scenarios: face in normal front, wearing hats and glasses, smiling or grimacing and low light’s environments.