Emotion Recognition Addon

Emotion Recognition Addon Link to heading

Real-time Emotion Detection for Video Conferencing

Overview Link to heading

This academic project presents the design and implementation of a machine learning model for recognizing human emotions from continuous video streams. The ultimate goal was to create a tool that could be used in video conferencing platforms for real-time emotion analysis.

Developed as a university project in 2021, the system divides the problem into two stages: face detection and emotion classification.

Problem Decomposition Link to heading

  1. Face Detection - Locate the face within an image frame
  2. Emotion Classification - Classify the detected face into an emotion category

Technical Implementation Link to heading

Face Detection Link to heading

  • Cascade Classifier - Haar cascade for face localization
  • Preprocessing - Resize to 48px, Gaussian blur, grayscale conversion

Feature Extraction Link to heading

  • Local Binary Patterns (LBP) - Texture descriptor for facial features
  • Bag of Words (BoW) - Alternative approach studied in literature review

Classification Link to heading

ComponentTechnology
AlgorithmSVM/LinearSVC
OptimizationGrid Search CV
CategoriesPositive, Neutral, Negative

Real-time Integration Link to heading

The system was integrated with the IRC protocol for real-time result display during video sessions, demonstrating practical application in communication platforms.

Dataset Link to heading

Training used an extensive, commonly accepted dataset for emotion recognition:

  • Standardized facial expressions
  • Multiple subjects
  • Three-category classification (positive/neutral/negative)

Technologies Used Link to heading

TechnologyPurpose
PythonCore implementation
OpenCVFace detection, image processing
scikit-learnSVM, Grid Search, model evaluation
IRC ProtocolReal-time result streaming

Academic Context Link to heading

This project was developed as part of my studies at Aristotle University of Thessaloniki. It explored machine learning approaches to emotion recognition, comparing methods like LBP and Bag of Words, and implementing a practical solution using SVMs.

The research included a comprehensive literature review of existing approaches, detailed methodology documentation, and evaluation of results across all stages of development.