A computational approach to screen scratch detection and analysis for improved maintenance of LCD terminals

Abstract

Manufacturers of credit and debit card terminals lend their products to retailers, retaining responsibility for repairing or replacing devices upon verification of malfunctions or cosmetic defects. The logistic operator, which processes up to ten thousand products daily, must assess each device’s condition through tests and manual visual inspections. Depending on the evaluation, a terminal may be returned to the client or sent to the manufacturer’s tech center for repair. This inspection process is highly subjective and prone to human error, as different inspectors may assess the same device inconsistently. To address these inconsistencies, we developed a visual inspection system specifically for evaluating the LCD conditions of payment terminals. The hardware includes a 3D-printed structure to optimize image acquisition, an active illumination scheme, and a camera to capture images of the terminals. The web application features a YOLOv8 object detection model to highlight defects on the terminal’s screen, complemented by a user-friendly interface with various functionalities. Our system significantly improved the efficiency and accuracy of inspections, providing operators with a reliable tool that ensures consistency. The final model achieved a mean Average Precision (mAP) of 77.4% and a processing speed of 88 milliseconds on CPU. Additionally, the real-time application interface offers low-latency usability, further enhancing the inspection process. After thorough model training and testing conducted in collaboration with the terminal manufacturer, the system has demonstrated its reliability for deployment at the production level.