Skip to content

Unleashing the Power of CNNs for Precise American Sign Language Recognition.

Notifications You must be signed in to change notification settings

saikeerthana-narra/American-Sign-Language-Detection-CNN

Repository files navigation

American Sign Language Detection with Convolutional Neural Networks

Unleashing the Power of CNNs for Precise American Sign Language Recognition.

This project focuses on training a convolutional neural network (CNN) model to recognize American Sign Language (ASL) letters. The goal is to classify hand gestures representing the letters A-Z (excluding J and Z) using a dataset of grayscale images.

DATASET

The dataset used for training and testing the model is similar in format to the classic MNIST dataset.

It consists of 27,455 training cases and 7,172 test cases, with each case representing a 28x28 pixel image and a corresponding label.

GETTING STARTED

To use this code, follow the steps below:

Clone the repository:

git clone https://github.com/saikeerthana-narra/American-Sign-Language-Detection-with-CNN.git

Install the required libraries

Download the datasets

Place the downloaded CSV files in the project directory

Run the code

MODEL ARCHITECTURE

The CNN model used for this project has the following architecture:

a) Convolutional layers with ReLU activation.

b) Max pooling layers for downsampling.

c) Batch normalization layers for normalization.

d) Dropout layers for regularization.

e) Fully connected layers with ReLU activation.

f) Softmax activation for the output layer.

g) The model is trained using the training data and evaluated on the test data.

h) The training process includes data augmentation techniques such as rotation, zooming, and shifts to improve generalization.

RESULTS

After training the model, the performance on the test data is evaluated. The following results were obtained:

LOSS: 0.04501

ACCURACY: 0.9898

In this case, the model is achieving a relatively low loss value and a high test accuracy.

A loss of 0.2815 indicates that the model's predictions have a small deviation from the true values.

A test accuracy of 0.9826 indicates that the model is correctly classifying or predicting approximately 98.26% of the samples in the test dataset.

These values suggest that the model is performing well and achieving a high level of accuracy on the given task.

FURTHER IMPROVEMENTS

Possible improvements for this project include:

Experimenting with different architectures or hyperparameters to improve model performance.

Exploring other data augmentation techniques.

Training the model on a larger dataset for better generalization.

Releases

No releases published

Packages

No packages published