A Comprehensive Analysis of VGG16 and ResNet50 Models in Sign Language Recognition

Sign language recognition is an essential aspect of facilitating effective communication between the deaf community and the rest of the population. The paper describes a method for addressing the problem of sign language misinterpretation by utilizing machine learning models. We proposed a convolutional neural network (CNN) - based sign language recognition system using pre-trained models such as VGG16 and Resnet50. The models are trained using a dataset that contains 26 ASL alphabets and numbers ranging from 1 to 10. Label encoding and one hot encoding are used to convert categorical labels into numerical representations, which are required for training ML models. A dense layer with dropout is applied to the classification head to reduce overfitting. When compared to Resnet50, the VGG16 architecture with customized layers for sign language identification performs better. The VGG16 model attained an accuracy of 99.92% while the Resnet50 model attained an accuracy of 99.47%. The outcomes demonstrate improved accuracy since we used a larger dataset to train the model than previously known datasets.