<![CDATA[Manual coin counting and classification in high-volume transactions are often inefficient and prone to human error. In computer vision implementations, the primary challenges in coin detection are lighting variations and changes in image scale, rendering conventional area-based methods inaccurate. This study aims to develop a coin detection system robust to changes in camera distance and lighting conditions. The proposed method combines the Hough Circle Transform algorithm for detecting geometric coin locations and ORB (Oriented FAST and Rotated BRIEF) for classifying denominations based on surface texture feature matching. The system is equipped with an adaptive learning mechanism (Human-in-the-Loop), allowing users to interactively train the system upon encountering new coin variants or unrecognized coins. Test results indicate that the combination of CLAHE (Contrast-Limited Adaptive Histogram Equalization) pre-processing and Hough detection successfully isolates overlapping circular objects, while the feature matching method effectively distinguishes coin denominations with similar diameters but distinct textures. It is concluded that integrating texture analysis with user manual correction features significantly enhances system accuracy and flexibility compared to static contour-based detection methods.]]>
Copyrights © 2026
