Multimodal Image Registration of Electrical Equipment Based on Feature Triangle

Infrared and visible imaging are crucial for detecting faults in electrical equipment, but their differences in resolution, field of view, and spectrum pose challenges for accurate image registration. Current methods often suffer from poor accuracy, lengthy processing times, or failing to register properly. This article proposes a multimodal image registration method for electrical equipment using feature triangles. It categorizes contours into open and closed contours, computes point curvatures to identify features, constructs feature triangles based on curvature radii, and determines principal directions using triangle centroids. The method demonstrates invariant properties under scaling, translation, and rotation. We evaluated the proposed method using both publicly available datasets and our own datasets, conducting both subjective and objective comparisons with state-of-the-art techniques. The proposed method achieved an average registration accuracy of 0.1117 and an average processing time of 8.834 s. For the infrared and visible images, the significant curvature contour ratios were measured at 0.251 and 0.193, respectively, while the feature point ratios were found to be 0.610 and 0.527. The experimental results demonstrate superior registration accuracy and reduced computational overhead compared to existing methods.