😊 Real-Time Emotion Detection

A real-time emotion recognition system using a Convolutional Neural Network (CNN) trained on the FER2013 dataset. Detects 7 emotions live from a webcam feed using OpenCV for face detection.

Results

Metric	Score
Test Accuracy	66.59%
Best Epoch	46
Dataset	FER2013 (35,000+ images)
Emotions	7 classes

Per-Class Performance

Emotion	Precision	Recall	F1-Score
😊 Happy	0.87	0.84	0.86
😲 Surprise	0.76	0.83	0.80
🤢 Disgust	0.75	0.56	0.64
😠 Angry	0.57	0.59	0.58
😐 Neutral	0.62	0.63	0.62
😢 Sad	0.53	0.56	0.55
😨 Fear	0.54	0.48	0.51

Features

• Real-time emotion detection from webcam
• Detects and tracks faces automatically using OpenCV
• Supports 7 primary emotions
• Runs entirely on local hardware — no cloud or API required
• Trained on FER2013 (35,000+ labelled facial images)

Tech Stack

Layer	Technology
Deep Learning	TensorFlow / Keras
Computer Vision	OpenCV
Model Type	Convolutional Neural Network (CNN)
Dataset	FER2013 (Kaggle)
Language	Python

How to Run

1. Clone the repo

git clone https://github.com/manny2341/Emotion-Detection.git
cd Emotion-Detection

2. Create a virtual environment

python3 -m venv venv
source venv/bin/activate

3. Install dependencies

pip install -r requirements.txt

4. Download the FER2013 dataset

Download from Kaggle and place it in the project folder.

5. Process the data

python3 data_processing.py

6. Train the model

python3 train_model.py

7. Run live emotion detection

python3 emotion_detection.py

Press Q to quit the webcam window.

Project Structure

Emotion-Detection/
├── data_processing.py        # Prepares and processes FER2013 dataset
├── train_model.py            # Builds and trains the CNN model
├── emotion_detection.py      # Live webcam emotion detection
├── Emotion_Detection.h5      # Saved trained model
├── training_history.png      # Accuracy and loss curves
├── haarcascade_frontalface_default.xml  # OpenCV face detector
└── requirements.txt

How It Works

1. OpenCV captures video from the webcam frame by frame
2. Haar Cascade detects face regions in each frame
3. Detected face is resized to 48×48 pixels and converted to grayscale
4. CNN predicts the emotion from the face image
5. Emotion label and confidence are displayed on the video feed in real time

Author

@manny2341