Dynamic mexican sign language recognition using LSTM networks for emergency response

Abstract

The integration of disciplines such as mathematics, numerical methods, computing, and mathematical modeling has driven the development of technological tools capable of identifying patterns and predicting phenomena with high accuracy. Among these tools, artificial intelligence has generated innovative solutions across several industrial and engineering sectors; however, its application aimed at supporting vulnerable populations remains limited. In this context, the use of artificial intelligence is explored as a support tool for the early detection of risk situations affecting the deaf community that uses Mexican Sign Language. The objective is to develop and evaluate a model based on Long Short-Term Memory recurrent neural networks capable of recognizing in real time Mexican Sign Language signs associated with emergency contexts. The proposed methodology is based on a recurrent neural network trained for real-time recognition of a set of signs related to emergency situations. The system processes video sequences frame by frame, identifies temporal patterns in hand, face, and body movements, and estimates the probability of occurrence of words linked to risk situations. The model achieved an accuracy of up to 100% in the identification of critical signs in the training and validation datasets. However, during testing, errors were identified in the recognition of signs with high gestural similarity. Despite the current limitations of the system, such as inference latency and the small dataset size, the results demonstrate the potential of this approach as a support tool for the early identification of risk situations. Furthermore, opportunities for future improvement are identified, including optimizing processing speed, expanding the recognized vocabulary, and advancing toward operational implementation in real-world environments.