ASL Recognition using Deep Learning Algorithms

There is a considerably vast communication gap between deaf and hearing individuals, and one potential solution is the development of American Sign Language (ASL) recognition technology. While methods utilizing deep learning and convolutional neural networks (CNNs) have shown promising results in general ASL recognition tasks, the potential of optimized CNN architectures for ASL recognition remains untapped. This research identifies the ideal CNN structure to achieve best ASL recognition in static images. The VGG16 model has come out to be very effective; however, analyzing multiple CNN implementations may provide interesting perspectives which could in turn expose features that could potentially be helpful in ASL recognition. This experiment compares the performance of a fine-tuned ResNet50 CNN model with a baseline VGG16 model in ASL recognition. Through analysis of each model’s respective strengths and weaknesses, the study aims to identify areas within the VGG16 architecture that may be further optimized to produce precise results. This research project provides guidance for picking suitable CNN architectures and offers valuable insights for improving VGG16 model to achieve even better accuracy and robust results.