Optical fiber sensor based on macrobends assisted by machine learning methods for recognizing static signs of the hand language alphabet

Abstract Purpose: A flexible optical fiber sensor assisted by machine learning was developed and applied on the recognition of hand static poses associated with the Brazilian language alphabet. The sensor stands out for its ease of manufacturing and interrogation, and may help to improve communication between speakers and non-speakers of sign language. Methods: The macrobends sensor comprises a single segment of a wave-structured optical fiber encapsulated in a flat hand shape made of silicone. Macrobends distributed along the fingers are simultaneously monitored through the transmission spectrum in the 400–800 nm range. Modifications in the geometry of the flat hand change the guiding conditions and consequently the transmitted light spectrum. Predictive models estimate hand gestures based on pose-dependent spectral features of the captured spectrum. A dataset containing 4000 transmission spectra (100 per pose $$\times $$ 10 poses $$\times $$ 4 days) was used to train and test the models. Results: Tests under repeatability conditions provided a maximum standard deviation of 4.7% in transmittance, reflecting the sensor’s ability to reproduce the same optical signal. The gestures associated with 10 letters of the Brazilian sign language alphabet were recognized by the Support vector classifier model with an accuracy of 97%. Conclusion: This work demonstrates the feasibility of the flexible sensor based on optical fiber macrobends for manual gesture recognition. The ease of installation in gloves and robotic hands makes the sensor a promising tool for application in areas such as sign language communication, rehabilitation engineering, and robotics.