Translation of Air-Written Baybayin Using Optical Flow in Complex Background

Motion tracking plays a vital role in computer vision, yet its use in real-time script translation, particularly in complex environments, remains underexplored. This study compares the effectiveness of two optical flow methods—Lucas-Kanade and Farneback—for real-time translation of air-written Baybayin characters in challenging backgrounds. The system utilizes Color Thresholding for background subtraction, optical flow for motion tracking, and a combination of Tesseract OCR with a Gated Recurrent Unit (GRU) model for character translation. Results show that the Lucas-Kanade method achieved a significantly higher accuracy of 82%, outperforming Farneback's 44%, which struggled with delayed tracking and incomplete character formation due to its higher computational demands. Lucas-Kanade's smoother and more reliable motion tracking allowed for more accurate character recognition, even in environments with complex backgrounds and variable lighting conditions. These findings demonstrate that Lucas-Kanade is the more effective optical flow method for real-time Baybayin translation, making it a promising approach for future applications in script recognition and translation tasks in dynamic environments.