Feedback System Usign The Humans Emotions

Preview:

Citation preview

1

LiveFeedback Penguins FCI-CU

Computer vision

2

Prepared by Moustafa Mohamed Ali Yasmin Abobakr Mo’men Mohamed Radwa Samy Tariq Senosy

3

Outline Introduction Problem Statement Background Suggested Solution Related work Project Relevance System Architecture Accuracy Dataset Demo Future work Questions

4

Introduction Emotions in everyday human

communication Communication ways :

by language : 7% by paralanguage : 38% by facial expression : 55%

5

Problem Statement Getting feedback about a specific

session from attendees during the session.

We can’t stop the session every minutes and ask the attendees about their opinion.

We need to find a way that’s Quick Effective Get good estimates

6

Background Computer Vision Systems Emotion Recognition techniques

Brain signals (EEG) Speech Facial Expression

7

Suggested Solution A computer vision system that can get

feedback from audience of a session by detecting their emotions –in real time- through Facial Expression Analysis.

Theme :

8

Related work AFFDEX android and web application MIT Learning Companion

9

Value Proposition Using such system will be valuable in

Learning purposes Speakers’ presentations Ads feedback

10

Project Relevance Software Engineering (Agile, VCS, SDLC) Genetic Algorithms Machine Learning Computer Vision

11

System Architecture

12

Feature Extraction

Feature Selection

Classification

Face Detection

13

Face Detection Viola Jones Algorithm (OpenCV library)

It's made for frontal face positions

14

Feature Extraction MethodsGabor FilterConvolutional Neural Nets.Local Binary Pattern Active Appearance Model

Chosen Method(s)Gabor Filter

15

Feature Reduction MethodsGenetic AlgorithmDown samplingPCAAdaboost

Chosen Method(s)Genetic Algorithm (Tested)Downsampling (Currently in use)

16

ClassificationMethodsSVM (lib-svm)Neural Networks

Chosen Method(s)SVM

17

Accuracy Validation 67.95% Testing 53.125% Factors affecting the accuracy

Dataset size Dataset variation Features Features normalization

18

Data set

With more than 1000 image created from different datasets we trained our system Japanese women database (213 image) 10k US Adult Faces (10,000 faces) Total number of images (1500 images)

19

Demo

20

Future Work Improve accuracy by trying different

methods for each phase Use datasets specified for feedback

emotions Android version

21

22

Thank You

Recommended