Video Emotion Analysis Toolkit

A robust video emotion analysis system combining OpenCV's Haar cascades with DeepFace's deep learning models for efficient facial emotion recognition in video streams.

Key Features

• Hybrid Face Detection: Combines OpenCV Haar cascades with DeepFace verification
• Adaptive Frame Processing: Skips frames (configurable) to optimize performance
• Face Preprocessing Pipeline: Automatic contrast/brightness adjustment + resizing
• Confidence-based Filtering: Ignores low-confidence predictions (<0.8 threshold)
• Temporal Aggregation: Groups results by second for stable emotion reporting
• Error Resilience: Comprehensive exception handling at all processing stages

Installation

pip install opencv-python-headless deepface numpy

Usage

python app.py <video_path>

Example Analysis:

python app.py demo_video.mp4

Frame: 15 Emotion: happy (Confidence: 0.92)
Frame: 30 Emotion: neutral (Confidence: 0.85)
Second 1: Emotion changed from happy to neutral

Technical Implementation

Processing Pipeline

1. Frame Decoding (OpenCV VideoCapture)
2. Hybrid Face Detection (Haar Cascade + DeepFace verification)
3. Face Normalization (Contrast/Brightness adjustment + Resize to 224x224)
4. Emotion Analysis (DeepFace's emotion model)
5. Temporal Aggregation (Per-second emotion statistics)

Configuration Constants

FRAME_SKIP = 5               # Process every 5th frame
EMOTION_CONFIDENCE_THRESHOLD = 0.8  # Minimum confidence score
EMOTION_CHANGE_THRESHOLD = 0.3      # Relative change for emotion shift

Optimizations

• Selective Frame Processing: Reduces redundant computations
• Face Preprocessing: Standardizes input for better model accuracy
• Memory Management: Explicit video capture release post-processing
• Confidence Filtering: Eliminates uncertain predictions

License

MIT License - See LICENSE for full text.

Contribution

Issues and PRs welcome! Please follow standard GitHub workflows.