Head Pose Estimation

This project implements Head Pose Estimation using machine learning and computer vision.

It allows predicting yaw, pitch, and roll angles from images, videos, and real-time webcam streams.

How It Works

Face Landmark Detection → Extract 3D facial landmarks using MediaPipe.

Preprocessing → Normalize and prepare features.

Machine Learning → Train Random Forest model on extracted features.

Prediction → Predict yaw, pitch, roll angles.

Visualization → Draw arrows on the face showing head orientation.

Pipeline Diagram:

Image/Video → Preprocessing → ML Model → Pose Angles → Visualization

Results

Metric Value Mean Absolute Error (MAE) ~5.4° Tested On AFLW2000 Dataset

Predictions are smoothed for stable video output.

Arrows represent yaw (left/right), pitch (up/down), and roll (tilt).