Context:

This article presents the design of a computational predictive model that facilitates the recognition of Colombian Sign Language (LSC) in a hotel and tourism environment.

Method:

Artificial intelligence techniques and deep neural networks were applied in the learning and prediction of gestures in real time, which allowed the construction of a tool to reduce the gap and strengthen communication. Convolutional neural network algorithms were applied to real-time data capture. Movement was captured using mobile device video cameras, thus obtaining the images that make up the data set. The images were used as training data for an optimal computational model that can predict the meaning of a newly presented image.

Results:

The performance of the model was evaluated using categorical measures and comparing different configurations for the neural network. In addition to this, everything is supported with the use of tools such as Tensorflow, OpenCV and MediaPipe.

Conclusions:

Finally, a model capable of identifying and translating 39 different signs between words, numbers and basic phrases focused on the hotel sector was obtained, where a success rate of 97.6% was obtained in a controlled use environment.

Acknowledgements:

Pedagogical and Technological University of Colombia - UPTC