Development of an End-to-End Form Data Capture Model for an Electronic Election Recapitulation System

• Main Authors: Chandra, James, Perdana, Riza Satria, Akbar, Saiful
• Format: Conference Proceeding
• Language: English
• Subjects: Adaptation models; AprilTag; Computational modeling; convolutional neural network; Data models; fiducial marker; image pre-processing; Optical character recognition; Optical imaging; optical mark recognition; Predictive models; Voting
• ISSN: 2640-0227
• DOI: 10.1109/ICoDSE59534.2023.10291866

This paper discusses the development of a data capture model that aids vote tabulation in the Indonesian election process. The model can scan and transcribe the standardized C-form into a digital format to reduce the potential for human error in manual transcription. The focus of the study is on building the end-to-end vision engine for Optical Character Recognition (OCR) and Optical Mark Recognition (OMR), creating a dynamic, format-adaptive model, and developing a form sheet page identification system to safeguard against mis-ordered page uploads. Various alternative methods are explored to address these key areas, such as alternative image preprocessing, testing OCR libraries versus self-built OCR using convolutional neural networks, implementing configuration files for engine dynamism, and fiducial marker identification. The self-developed OCR model succeeded in detecting digits 0-9, achieving an average accuracy of 89% per number. Furthermore, dynamization efforts with configuration files and contour detection had an average computation time of just 0.0021 seconds to find the field location on the second page sample of the form. The page identification feature was successfully implemented, utilizing the AprilTag fiducial marker due to its resistance to occlusion, fast computation time, and high reading accuracy.