Real-Time Emotion Recognition Using Facial Expressions

This repository contains the code for a real-time emotion recognition system developed using OpenCV and TensorFlow. The project aims to detect and classify human emotions from live video input captured via a webcam, leveraging both traditional computer vision techniques and deep learning models.

Project Overview

This project focuses on creating a robust emotion detection system that can process live video input, detect faces, and predict emotional states using a deep learning model. The key components of this system include:

• OpenCV: Used for real-time face detection via Haar Cascades.
• TensorFlow: A pre-trained deep learning model classifies emotions into seven categories: Angry, Disgust, Fear, Happy, Neutral, Sad, and Surprise.

Methodology

The system integrates two major components:

1. Face Detection: Using Haar Cascade classifiers to detect faces from the live video feed.
2. Emotion Classification: Leveraging a convolutional neural network (CNN) built with TensorFlow to classify the detected facial expressions into emotional categories.

Key Features:

• Real-time processing of live video input.
• Emotion classification with high accuracy using deep learning techniques.
• Modular and scalable code architecture, making it easy to extend or modify.

Technologies Used

• OpenCV: For real-time face detection.
• TensorFlow/Keras: For deep learning and emotion classification.
• Python: The core programming language used for implementation.

Setup and Installation

1. Clone the repository:

   git clone https://github.com/yourusername/real-time-emotion-recognition.git
   

2. Install the required dependencies:

pip install opencv-python-headless pip install tensorflow pip install numpy pip install matplotlib pip install seaborn pip install pandas

3. Run the emotion detection script: python emotion_detector.py

Usage

The system captures video input from your webcam, detects faces, and then classifies emotions in real-time. The deep learning model processes grayscale images of size 48x48 pixels and classifies them into one of the seven emotions.

Model Details

The model was trained on the Face Expression Recognition Dataset from Kaggle and consists of the following layers:

Convolutional layers for feature extraction. Max Pooling layers to reduce dimensionality. Dropout layers to prevent overfitting. Dense layers for classification. The model achieved an accuracy of 77.8% on the test set, with particularly strong performance in detecting emotions such as 'Happiness' and 'Surprise.'

Results

Training Accuracy: 77.8% Validation Accuracy: Varies, with some fluctuations due to potential overfitting. Precision and Recall: Strongest for emotions like 'Happiness' and 'Surprise', while emotions like 'Disgust' and 'Fear' showed lower detection rates.

Future Work

Further improvements could be made by:

Training with a more diverse dataset to improve accuracy across all emotions. Experimenting with more advanced neural network architectures. Extending the model to handle varying cultural expressions and lighting conditions.