基于像素角点检测的无人机测绘多源遥感影像自动配准技术

Automatic registration technology for multi-source remote sensing images in unmanned aerial vehicle surveying based on pixel corner detection

摘要:

在无人机测绘多源遥感影像自动配准过程中,平移差过大会导致无人机测绘多源遥感影像的不匹配对接,且无人机测绘多源遥感影像中存在不适合用于配准的角点,导致影像自动配准的精度较差。为解决上述问题,设计基于像素角点检测的无人机测绘多源遥感影像自动配准技术。定义像素信息的多尺度空间,完成多源像素匹配,推导多尺度特征模型表达式,提取关键像素信息,实现多源遥感影像像素信息取样。预处理遥感图像,检测像素角点,通过去除非配准角点的处理方式,确定配准处理主方向,再按照细节增强标准,完善具体的配准操作流程,完成基于像素角点检测的无人机测绘多源遥感影像自动配准技术的设计。实验结果表明,应用所提方法后,遥感图像像素的平移差保持在0-35pt的数值范围内,在像素采集尺度不唯一的情况下,有效解决了由平移差过大导致的无人机测绘多源遥感影像不匹配对接的问题,图像配准重叠率较高,保障了配准后图像的真实性。

Abstract:

In the process of automatic registration of multi-source remote sensing images in drone surveying, excessive translation errors can lead to mismatched docking of drone surveying multi-source remote sensing images, and there are corners in drone surveying multi-source remote sensing images that are not suitable for registration, resulting in poor accuracy of image automatic registration. To address the above issues, a multi source remote sensing image automatic registration technology for unmanned aerial vehicle mapping based on pixel corner detection is designed. Define a multi-scale space for pixel information, complete multi-source pixel matching, derive multi-scale feature model expressions, extract key pixel information, and achieve multi-source remote sensing image pixel information sampling. Preprocess remote sensing images, detect pixel corners, and determine the main direction of registration processing by removing non registration corners. Then, according to the detail enhancement standards, improve the specific registration operation process, and complete the design of automatic registration technology for multi-source remote sensing images in unmanned aerial vehicle surveying based on pixel corner detection. The experimental results show that after applying the proposed method, the translation difference of remote sensing image pixels is maintained within the numerical range of 0-35pt. In the case of non unique pixel acquisition scales, the problem of mismatched docking of multi-source remote sensing images in unmanned aerial vehicle mapping caused by excessive translation difference is effectively solved. The image registration overlap rate is high, ensuring the authenticity of the registered image.