Project Work Genika

This is the project repository for Genika. The Team is made up by:


Carla Brugulat	Javier de la Rica	Montse Sanahuja	Joan Sansa	Christian Segovia	Alex Kim	Chao-Feng Wang

It is going to be explained below what Genika has been done during the PAE course at UPC at Spring 2017.

The main aim of the current project is to achieve a real time facial detection in order to create a personalized avatar the user would be able to play with.

Stage 1 - Android

The team computed an Android Application that has access to the main and the frontal camera. The App has also a GET AVATAR button, which takes a picture and saves it, so later it computes the avatar and shows it on the device's screen. The team has also computed the main page of the Android Application with a simple Login page, a button to directly open the frontal camera and a link to Genika's website with the logo.

Stage 2 - Face Detection

Genika has been able to execute the Dlib C++ library. Dlib is a C++ toolkit containing machine learning algorithms complex software. The team used it to detect the user's face, and observed that it detects different faces in a single image in a prevcise way. Furthermore, Genika has tried several images with different characteristics in order to see the strength of the algorithm, studying at the same time the Dlib code. The system detects the faces even if they are dark, and even if the person wars glasses or the picture is blurry or Black and White.

Results

				   From 92 desirable faces to detect:
					- 80 faces detected.
					- 12 faces not detected.

				    | ACCURACY = 80/92 = 86.95% |

Figure 1. Percentages of the Dlib and App tests.

The system also extracts a .txt file with the landmarks of each point of the face detection. Genika worked on a simple Matlab code studying the way to modify the presentation of these landmarks and make a proper use of them, so later it would be used in the 3D Meshing stage.

Figure 2. Landmarks detection and classification.

Stage 3 - 3D Meshing

From the 68 landmarks detected in the images, the team wants to extract these points and place them in a 3D Model in such a way that the image fits into it. The team has designed a 3D model using Blender. Blender is an open source 3D creation suite that supports the entirety of the 3D pipeline-modeling, animation and simulation. Genika computed an Android Application which shows the 3D model and it allows to load the images into the model.

Figure 3. 3D Meshing and basic 3D Model.

Stage 4 - Integration

Genika computed an Android Application in which all the software works for the same Application. This Application has been tested in different Android devices in order to test the proper functionality and all the code is available:

Application Code:

Face Detection App

Research Code: