The development of a humanoid robot includes 6 degrees of freedom and advanced computer vision algorithms.
Project objectives
-Design a humanoid to test artificial vision algorithms.
-Simple and economical elements.
-Control 6 axes of independent movements.
-Use the face of a humanoid robot Open source.
-Write a stepwise algorithm for face, pupil and mouth recognition using different techniques and with the use of image processing and preprocessing methods.
-Integrate an algorithm for tracking to the detection methods implemented.
-To make the humanoid mimic the user's movements.
-Versatile and intuitive user interface.
-Visualization of the image properties captured by the algorithm.
Facial recognition
These algorithms were developed from the ground up in MatLab, without the use of libraries, using MatLab's ccaracteristics of a Viola-Jones-based recognition.
Robust - High true positive detection and low false positive and false positive performance real time.
4 stages of the algorithm:
Binary image.
Search with Adaboost.
Cascade classifiers
Haar-Features.
HAAR-FEATURES
Creation of an integral image-> V-J
ROI
Sum of pixels per zone in each window.
Processing speed.
ADABOOST
Predictive algorithm for classification developed by matlab.
Prepared to prevent overfitting.
Depending on the model, it is considered adaptive because it is composed of simpler classifiers.
The result is a classifier that is more accurate than learning algorithms.
This project was carried out as part of the TFM in the master of automation and robotics, using the connection of an Android cell phone and an application developed for the control and acquisition of images and a system of execution of movements printed in 3D, driven by an Arduino and an active connection by BUS with MatLab.