GeReco

Vision & Cognitive Systems Project

Authors:

• Andrea Auletta - 2107158 - andrea.auletta@studenti.unipd.it
• Marco Bernardi - 2107781 - marco.bernardi.11@studenti.unipd.it
• Sebastiano Sanson - 2130917 - sebastiano.sanson@studenti.unipd.it

---

Project Overview

GeReco (Gesture Recognition) is a project focused on recognizing both dynamic and static gestures using deep learning techniques, including LSTMs and Mediapipe-based models. The system is designed for real-time inference using a standard USB camera.

---

Directory Structure

dynamic-hg/

Contains the implementation for dynamic gesture recognition using an LSTM-based approach with dimensionality reduction, inspired by Next-Gen Dynamic Hand Gesture Recognition: MediaPipe, Inception-v3 and LSTM-Based Enhanced Deep Learning Model.

• inference.ipynb: Notebook for performing real-time inference using a trained model and a USB camera.
• training.ipynb: Notebook for data preprocessing and training the LSTM model.

static-hg/

Contains the implementation for static gesture recognition using Mediapipe.

• inference.py: Script for real-time inference of static gestures.
  • Usage: python inference.py <model-path> <OpenCV-Device-Id>
• training.ipynb: Notebook for data preprocessing and training the model.

results/

Stores snapshots of training results and performance metrics.

• MIL_results.md: (Mediapipe Inception LSTM) Contains logs of training results from multiple runs.
• MP_results.md: (Mediapipe) Contains logs of training results from multiple runs.

paper/

Contains the source files for the project paper and the compiled pdf.

License

This project is licensed under the MIT License.