Facial Expression Classification Using ARKit Blendshape Time-Series Data
Facial Expression Classification Using ARKit Blendshape Time-Series Data
Project Overview
This project presents a lightweight facial expression recognition pipeline that classifies seven Ekman-style emotions from ARKit facial motion-capture time series. Using Edge Impulse and a Random Forest model trained on 52 blendshapes plus head/eye pose data, the system achieves 100% validation accuracy and 70% sequence-level accuracy, with potential for real-time MetaHuman emotion recognition in Unreal Engine.
Video Demonstration
Check out this video demonstration showing the facial expression classification system in action, with real-time emotion recognition from facial motion-capture data.