Hand Gesture Recognition System

A computer vision project that recognizes hand gestures using TensorFlow and OpenCV, with optional Raspberry Pi GPIO integration.

Overview

This project combines deep learning, computer vision, and IoT technologies to create a contactless home control system. Using a camera and trained CNN model, it recognizes eight distinct hand gestures to control different home appliances simulated through GPIO devices on a Raspberry Pi.

Key Features

• Real-time hand gesture recognition
• CNN-based gesture classification
• GPIO-controlled device simulation
• LCD status display
• Support for both desktop and Raspberry Pi environments
• Custom dataset creation and model training capabilities

Project Structure

The project contains the following key files:

Data Collection

• getimage_real.py: Captures hand gesture images from webcam for creating the dataset. Images are saved in grayscale format.

Model Training

• train_model.py: Trains the CNN model using TensorFlow/Keras with the collected dataset.
• tflite_convert.py: Converts the trained Keras model (.h5) to TensorFlow Lite format for embedded devices.

Model Evaluation

• result_test.py: Evaluates model performance on test dataset, generates classification metrics and confusion matrix.
• result_validation.py: Evaluates model performance on validation dataset.
• test_image.py: Tests the model on a single image file.

Implementation

• verify_webcam.py: Real-time hand gesture recognition using webcam (desktop version).
• verify_gpio.py: Hand gesture recognition with GPIO control (Raspberry Pi version).
• recognise_gpio_cam.py: Real-time hand gesture recognition with GPIO control (Raspberry Pi version).
• recognise_gpio_img.py: Image-based hand gesture recognition with GPIO control (Raspberry Pi version).

Setup Instructions

1. Install required packages:

pip install -r requirements.txt

2. For Raspberry Pi GPIO functionality:

• Enable GPIO in Raspberry Pi configuration
• Connect LCD display and GPIO components according to pin configuration
• Run with sudo privileges for GPIO access

Usage

1. Collect Dataset:

python getimage_real.py

2. Train Model:

python train_model.py

3. Run Recognition:

• For desktop webcam:

python verify_webcam.py

• For Raspberry Pi:

sudo python recognise_gpio_cam.py

Supported Gestures

The system recognizes 8 different hand gestures:

• FIST
• ONE
• TWO
• THREE
• THUMBSUP
• FIVE
• SIX
• SEVEN

GPIO Control (Raspberry Pi)

The system controls:

• 2 LEDs
• 1 Buzzer
• 2 Motor endpoints
• LCD Display

Each gesture triggers different combinations of these outputs.

Model Architecture

The CNN model consists of:

• 2 Convolutional layers
• 2 MaxPooling layers
• Dropout layers
• Dense layers
• Input shape: (64, 64, 1) - Grayscale images
• Output: 8 classes (gestures)

Dependencies

• TensorFlow
• OpenCV
• NumPy
• Imutils
• Scikit-learn
• Pandas
• Matplotlib
• Seaborn
• RPi.GPIO (for Raspberry Pi)
• rpi-lcd (for Raspberry Pi)