How adityakamat24/sign-language-detection works
Main Components
• Data Capture Module: Represented by the "Capture dataset.ipynb", this component handles image collection using a webcam.
• Data Preprocessing & Dataset Creation: Although not a separate file, preprocessing steps (resizing, grayscaling, normalizing) are part of the dataset preparation workflow.
• Training Module: "Training.ipynb" captures the neural network training phase using the preprocessed dataset. The trained model artifacts include "model-bw.h5" (weights) and "model-bw.json" (model architecture).
• Real-time Application Module: "application.ipynb" ties together real-time image capture, preprocessing, and inference using the pre-trained model. This component also incorporates the Tkinter GUI to interface with the user.
• Utility Component: "make folders.ipynb" could be included as a helper for setting up the expected directory structure (for saving datasets, models, etc.).
Relationships and Data Flow
• The flow begins with the Data Capture Module where images are taken from the webcam.
• The captured images then pass through the Data Preprocessing & Dataset Creation stage, where they are transformed (resized, converted to grayscale, normalized).
• The preprocessed dataset is used in the Training Module to build a neural network model (using TensorFlow), with output stored as model files.
• Finally, the Real-time Application performs continuous image acquisition through the webcam, processes each frame (using OpenCV), and applies the trained model to recognize signs. Results are displayed in the Tkinter-based GUI.
Architectural Patterns & Technologies
• The project follows a pipeline-based pattern where data flows sequentially through capture, processing, training, and inference stages.
• Key technologies: Python for scripting, OpenCV for computer vision, TensorFlow for deep learning, and Tkinter for the GUI.
• Separation of concerns: each notebook represents a distinct stage of the workflow ensuring modularity.