Enhanced Feature Matching: Entropy-Guided ORB with Adaptive Descriptors

In real-time computer vision applications, robust feature matching is crucial for environment perception and localization. Traditional ORB algorithms struggle under low-texture or significant illumination variations. We propose EAD-ORB, an enhanced algorithm incorporating an entropy-guided adaptive thresholding mechanism for keypoint detection and a dual-channel descriptor for improved adaptability. Experimental results on the HPatches dataset demonstrate an average matching accuracy of 83.20%, significantly outperforming ORB (80.18%) and other methods under illumination variations. EAD-ORB's training-free, stable, and geometrically accurate performance makes it suitable for complex image matching and SLAM applications.The code of this work is publicly available at:https://github.com/Mayrou/EAD-ORB