Enhancing American Sign Language Recognition: A Comparative Study of LSTM and TensorFlow Zoo Models for Static and Non-Static Gesture Detection

Language is the cornerstone for communication, enabling the exchange of information, ideas, culture, and emotions. Communication relies on motions and signs, collectively known as sign language for individuals with hearing and speech impairments. However, effective interaction between those who are familiar with sign language and those who are not often proves challenging due to limited awareness and reluctance to learn it. This study centers on the recognition of American Sign Language (ASL). A review of 16 research articles was conducted to explore various methods and compare their effectiveness. Most studies utilized static images for recognition, employing models based on CNN (Convolutional Neural Networks) or its variants, as well as SVM (Support Vector Machines) and KNN (K-Nearest Neighbors). In this study, the TensorFlow Object Detection Zoo model (SSD MobileNet V2 FPNLite 640x640) and LSTM (Long Short-Term Memory) are evaluated to determine the most effective technique for detecting dynamic and static ASL alphabets. Both models incorporate body movements and hand gestures during training, utilizing a self-developed dataset for evaluation.