A Fast Selection Method of Landmarks for Terrain Matching Navigation

面向地形匹配导航的地标点快速选取方法

¹ School of Automation, Northwestern Polytechnical University, Xi'an 710072, China
² College of Electronic and Information Engineering, Taiyuan University of Science and Technology, Taiyuan 030024, China

Received: 12 September 2019

Abstract

Under fixed imaging conditions, the landmark selection method based feature traversal analysis has high computational complexity. The hierarchical statistical significance detection method uses global statistical information for feature analysis to overcome the computational complexity problem caused by feature traversal analysis. The frequency domain de-correlation method can remove repeat mode in the image by adaptive Gaussian filtering on the amplitude-frequency characteristics. In this paper, combined the hierarchical statistical saliency detection method with the frequency domain de-correlation method, a fast landmark selection algorithm based on saliency analysis is proposed. Based on the proposed algorithm, the automatic landmark selection architecture for terrain matching navigation was constructed. The selection of landmark points was carried out in the Qin-ling Mountains and the Guangdong and Guangxi hills. The results show that compared with the feature or pixel-based landmark selection method, the landmark selection efficiency of the proposed method is improved by 2 to 3 orders of magnitude. The correct matching rate of candidate landmarks selected in the Qinling Mountains and the Guangdong and Guangxi hills are 73.9% and 88.3% respectively.

摘要

在固定成像条件下，对成像区域进行特征遍历分析的地标点选取方法存在计算复杂度高的问题。分层统计显著性检测方法采用全局统计信息进行特征分析可克服特征遍历分析带来的计算复杂度问题，频域去相关显著性检测方法通过对幅频特性的自适应高斯滤波可去除图像中的重复模式，通过融合分层统计显著性检测方法和频域去相关显著性检测方法，提出一种基于显著性分析的地标点快速选取算法。基于此算法构建了面向地形匹配导航的地标点自动选取架构，在秦岭山区和两广丘陵进行了地标点的选取实验。结果表明：新算法与基于特征或像素的地标点遍历分析方法相比，其地标点选取效率提升2至3个数量级，在秦岭山区及两广丘陵选取的候选地标点正确匹配率分别达到73.9%和88.3%。