ORB Feature Matching Algorithm Based on Multi-Scale Feature Description Fusion and Feature Point Mapping Error Correction

To improve the accuracy of feature extraction and description various scales in traditional Oriented FAST Rotated BRIEF (ORB) matching algorithm, this paper proposes an ORB algorithm based on multi-scale fusion point mapping error correction. Firstly, when establishing scale pyramid, method using same patch-size for each layer is adopted instead different patch-sizes a unified original which enhances robustness descriptor improves accuracy. Secondly, FAST-SCORE maps are established scales, coordinates high-level points mapped to bottom corrected further positioning points. The verified remote sensing images, autonomous driving, industrial automation fields. Experimental results show that resisting theoretical interference, average proposed 67.9%, about 2.0 times stability 14.0, 1.5 algorithm. After correcting mapping, can be improved by 19.2%, indicating has excellent interference. In experiments KITTI custom datasets, reached 88.70% 96.88%, respectively, improvement 10.15% 1.2% compared At time, time was reduced 17.34% 24.30%, ensuring real-time performance practical scenarios.