hand-gestures-recognition-and-segmentation-using-CNN-and-OpenCV

This GitHub repository contains code for a Convolutional Neural Network (CNN) model to detect five different hand gestures from segmented hand photos. The model is trained on a dataset using Keras with TensorFlow backend and includes data preprocessing, model building, training, and evaluation. Additionally, there is a separate file, "Real_life_test," demonstrating real-time hand gesture recognition using OpenCV.

Project Structure

The repository is organized as follows:

1. Data Preparation:

   • The dataset is stored in the "data" directory, with subdirectories for each class representing different hand gestures. You can download the dataset from this Kaggle link and extract it into the "data" directory.

2. CNN Model:

   • The initial model architecture is defined using Keras Sequential API in the main file.
   • The first attempt may show signs of overfitting, so a second model is created with dropout and regularization to address this issue.
   • The third model utilizes a pre-trained ResNet50 model for feature extraction, followed by additional layers for classification.

3. Training and Evaluation:

   • The models are trained using the training set and evaluated on the validation and test sets.
   • Training history, loss, and accuracy plots are visualized for each model.
   • Confusion matrices and classification reports provide detailed performance metrics.

4. Real-time Hand Gesture Recognition:

   • The "OpenCV_test" file demonstrates how to use OpenCV for real-time hand gesture recognition.
   • It captures video frames from the camera, preprocesses them, and utilizes the trained model to predict gestures.

5. Dataset Testing:

   • The "CNN02_Model_Test" file allows testing the trained model on any provided segmented hand photos.
   • Provide the directory path containing the segmented hand photos, and the model will predict the gestures.

Instructions:

1. Dataset:

   • Download the dataset from Kaggle and extract it into the "data" directory.

2. Training:

   • Run the "CNN_Train" file to train and evaluate the CNN models.
   • Run the "segmentation_ResNet(one real data)" file to Train and evaluate the ResNet model on the data without doing data augmentation
   • Run the "segmentation_ResNet(one augmented data)" file to Train and evaluate the ResNet model on the data after doing data augmentation
   • Experiment with model architectures and hyperparameters to achieve optimal performance.

3. Real-time Testing:

   • Execute the "OpenCV_test" file to see real-time hand gesture recognition using OpenCV.

4. Model Saving:

   • The trained models are saved in the repository for later use.

Requirements:

• Python 3.x
• Libraries: TensorFlow, Keras, OpenCV, scikit-learn, matplotlib, seaborn

Feel free to customize and extend the code according to your requirements. For any issues or suggestions, please create an issue in the repository.

Note: The trained ResNet models perform better than the CNN model

Happy coding! 🚀