This technology extracts Harris corners from visible and infrared images, categorizes them into subsets based on their gradient directions, and calculates a weighted Hausdorff distance by summing the Hausdorff distances of each subset. It is used as a similarity metric for multi-sensor image registration.
Cross-correlation-based registration is computationally complex, mutual information-based registration suffers from poor performance due to local brightness variations between sensors, and SIFT/SURF methods are vulnerable to deformation and noise because Euclidean distance lacks spatial information. Existing Hausdorff distance techniques also have limitations, including high computational load and the failure to account for gradient direction and the relative importance of each point.
This technology groups corners by gradient direction intervals to compare distances only between components with the same orientation. It applies weights based on corner response values and a cost function that filters out corners causing large distance errors, resulting in precise distance calculation. This improves the registration accuracy of visible and infrared images. It can be applied to surveillance cameras in nighttime or foggy conditions and thermal-visible fusion systems, enabling stable overlay and analysis of images from both sensors even in poor lighting.
N/A
