Anàlisi facial en entorns d’interacció home-màquina

Abstract

This project analyzes the most recent state of the art methods for face fitting in images as a first step of the methodologies required to be applied in human-machine interaction environments. The project centers its evaluation on the Explicit Shape Regression (ESR) and the Robust Cascade Pose Regression (RCPR) methods, which stem from a cascade regressors methodology with the objective of finding a set of points of interests of the faces in the images. This methodology of cascade of regressors is an initial requirement in most human-machine interaction scenarios; gender, age, ethnicity, face and emotion recognition, just to mention a few. To perform this analysis the public data base BOSHPORUS BD has been modified. The facial images of citizen with European ethnicity have been rotated and projected from different angular perspectives to carry out training of the previously appointed methods under different simulated conditions. The recent development of new methods for the detection of faces in images under different conditions requires researchers to analyze and compare these tools with the objective of determining which method offers the highest performance under different conditions of use as well as computational restrictions. As the result of this project two of the most frequently used methods of the state of the art have been evaluated and compared on the generated data to reach conclusions on the pros that make them the most suitable for different applications and the cons that open the door to future lines of research.