Implementasi Pencocokan Citra Menggunakan Metode ORB untuk Deteksi Logo Kemasan Produk

Abstract

This study evaluates the reliability of the Oriented FAST and Rotated BRIEF (ORB) algorithm for image matching and logo detection on product packaging in dynamic real-world environments. In modern industry, object detection is closely related to logo identification for automation. Testing was conducted through a series of systematic experiments covering variations in rotation, drastic scale changes, and background clutter. Pre-processing was done with grayscale conversion, followed by feature extraction using ORB, and matching using the Brute-Force Matcher with Hamming Distance. To filter out false matches, this algorithm was optimized with K-Nearest Neighbor (KNN), Lowe's Ratio Test, and RANSAC geometric filtering. Experimental results show that ORB optimization with KNN and RANSAC drastically reduces computation time to 0.2412 seconds and produces 51 valid inliers. However, despite being highly efficient for real-time applications, the ORB method remains fundamentally limited when facing extreme scale variance. Therefore, this system is optimally recommended only in controlled environments like factory conveyors. Alternative approaches using Deep Learning or scale-invariant algorithms (SIFT/SURF) are highly suggested for dynamic environment implementation.