Real-Time Hand Gesture Recognition System

This project was developed for the Machine Vision and Intelligence (MVI) course assignment. Our team built a vision-based system capable of detecting and classifying various hand gestures in real-time using advanced machine learning techniques.

Overview

The primary goal of this project is to create an intuitive interface for human-computer interaction through hand gestures. The system processes video input to accurately identify hand presence, orientation, and specific gestures like finger counting and swiping motions. We leverage state-of-the-art computer vision and deep learning models to achieve robust performance.

Key Features

• Real-Time Processing: Analyzes video streams to detect and classify gestures instantaneously.
• Finger Count Detection: Accurately counts the number of fingers extended (e.g., 1, 2, 3, 4, 5).
• Swipe Gesture Recognition: Detects directional swiping motions (e.g., left, right, up, down - confirm directions if applicable).
• Hand Orientation Classification: Differentiates between:
  • Left vs. Right Hand
  • Palm vs. Back of Hand
• Object Detection: Utilizes YOLOv8 for robust hand detection within the video frame.
• Gesture Classification: Employs Convolutional Neural Networks (CNNs) for accurate classification of detected hand gestures.

Technologies Used

• Programming Language: Python
• Computer Vision: OpenCV (for image preprocessing, video stream handling)
• Object Detection: YOLOv8
• Deep Learning / Classification: Convolutional Neural Networks (CNNs) - (Specify framework if possible, e.g., TensorFlow, PyTorch)

System Architecture (Conceptual)

1. Video Input: Capture video stream from a camera source.
2. Image Preprocessing: Use OpenCV for frame resizing, color conversion, and normalization.
3. Hand Detection: Apply the YOLOv8 model to detect the location (bounding box) of hands in the frame.
4. Region of Interest (ROI) Extraction: Crop the detected hand region for focused analysis.
5. Gesture Classification: Feed the cropped hand image into the trained CNN model to classify the gesture (finger count, swipe, orientation).
6. Output: Display the recognized gesture information on the video feed or output it for further application use.

Team Members

• Shokri Eyad Shokri Ouda (TP065881)
• Aravind Soundirarajan (TP066273)
• Pedro Fabian Owono Ondo Mangue (TP063251)

---