|
With the continuous development and application of multispectral technology, multispectral cameras play an increasingly important role in the field of camouflage target reconnaissance. By simultaneously obtaining image data from different bands, they provide richer spectral information for target recognition. In the imaging requirements of high-speed dynamic scenes, fast and accurate registration of multispectral images has become a key technical issue, directly affecting the imaging quality and practicality of the system. In response to this situation, this article proposes an improved ORB image registration algorithm. Firstly, a scale pyramid is constructed and the ORB algorithm is used to extract feature points. The BEBLID descriptor is used to describe the feature points, and the nearest neighbor ratio (NNDR) algorithm is used for coarse matching; Then, based on feature point voting, an optimized geometric constraint is constructed to further optimize the feature points. The Random Sampling Consistency (RANSAC) algorithm is used to calculate the transformation matrix and obtain a high-precision transformation matrix; Finally, affine transformation is used to achieve image registration. The experimental results show that the algorithm has high registration accuracy and low registration time. Compared with traditional algorithms, the proposed algorithm has higher registration efficiency and can complete high-precision image registration while reducing image registration time.
|
