Baybayin Translation Using Lucas-Kanade Optical Flow and RNN

Machine learning is continually evolving, leading to advancements in translating and interpreting various writing systems. Numerous studies have aimed to assist learners and researchers in this field, primarily focusing on image processing techniques to recognize and translate Baybayin characters from pre-written inputs. However, there is limited research on utilizing optical flow methods for capturing stroke movements in real-time handwriting applications. This study addresses this gap by investigating the effectiveness of the Lucas-Kanade optical flow method for tracking and translating drawn Baybayin characters into their respective scripts. The developed system captures user-drawn characters through a Raspberry Pi camera, employing the Lucas-Kanade algorithm to convert motion patterns into usable data inputs. The study integrates Optical Character Recognition (OCR) and a Recurrent Neural Network (RNN) to facilitate translation. A confusion matrix was used to evaluate the system's accuracy, resulting in an impressive rate of 87.5%. These findings indicate that Lucas-Kanade optical flow is a viable approach for translating single-stroke handwritten Baybayin characters.