AirSign: Real-Time ASL Recognition
PythonPyTorchYOLOv8OpenCVMediaPipe
AirSign is a real-time American Sign Language (ASL) recognition system that translates hand gestures into text. The project demonstrates a complete end-to-end machine learning workflow, from data preprocessing to real-time inference.
The Process
- Engineered a robust data pipeline to process the WLASL video dataset, using MediaPipe for intelligent frame extraction to create a high-quality, annotated image set for training.
- Trained and fine-tuned a custom YOLOv8 object detection model on this dataset.
- Achieved a high mean Average Precision (mAP50-95 of 0.79) on the challenging task of recognizing nuanced hand gestures.
- Built a real-time inference pipeline using OpenCV to capture webcam footage and display the translated sign language.