Curvature Scale Space Corner Detection-Based Registration Algorithm for Infrared and Visible Light Images of Power Equipment

To address the challenges of incomplete feature representation and insufficient localization accuracy in multimodal fault diagnosis of electrical equipment, this paper proposes a dual-modal feature co-registration algorithm termed Contour Angle Orientation -Speeded Up Robust Features (CAO-SURF). The methodology establishes a three-stage progressive feature enhancement framework through feature space fusion and adaptive registration mechanisms for multimodal sensing data. First, a Modal Perception-Based Canny Edge Detection Strategy (MP-Canny) is developed to resolve feature sparsity in low-contrast infrared images and noise interference in visible-light images. Subsequently, a contour angle orientation (CAO) and a Null-Value Area Compensation Strategy (NVACS)-based gradient field reconstruction model are developed to achieve rotation-invariant feature representation. Finally, an improved SURF descriptor integrated with bidirectional geometric constraint matching significantly enhances feature robustness. Systematic ablation experiments demonstrate the proposed method outperforms state-of-the-art approaches in core metrics including registration accuracy and recall rate. The innovatively developed feature space collaboration mechanism establishes a novel multimodal information fusion paradigm for electrical equipment condition monitoring.