A real-time video analysis system that detects whether a person is telling the truth or lying based on facial movements using Google's FaceMesh model. This research project was part of my university final year work focused on behavioral AI and facial recognition.

Project Development Steps:

• 📦 Imported and installed all required dependencies
• 🧠 Used Mediapipe's FaceMesh to extract 468 facial landmarks
• 📍 Collected keypoint values from facial data for training/testing
• 🗂️ Created labeled datasets for truth and lie expressions
• 📊 Preprocessed data into features and labels for model training
• 🔁 Built and trained an LSTM Neural Network using TensorFlow
• 📈 Evaluated model accuracy using confusion matrix
• 🎥 Integrated real-time video prediction and testing

Key Technologies Used:

• 📹 OpenCV – Live camera video feed
• 🧠 Mediapipe FaceMesh – Landmark extraction
• 📊 TensorFlow/Keras – Deep learning model (LSTM)
• 📁 Pandas/Numpy – Data preprocessing
• 📂 Matplotlib/Seaborn – Model evaluation visualization

This project helped me explore machine learning in video analysis and understand how human facial behavior can be mapped to AI predictions. It was a complete hands-on experience from data collection to real-time implementation, boosting my confidence in AI and deep learning fields.