摘要
【目的】从低空森林航摄影像特点出发,探索一种适合无人机森林航摄影像的分区匹配策略,以解决直接采用摄影测量软件处理低空无人机森林航摄影像时经常出现的匹配速度慢、错误匹配率高、匹配质量差等问题,为森林航摄影像处理相关研究提供参考。【方法】基于摄影测量和计算机视觉相关原理,提出一种借助无人机POS数据构建像对间单应约束、按一定位置分布规则划分若干对应子区域、在特征提取和配对过程中视每个子区域为独立影像的分区匹配策略;同时将GIS空间分析算法引入特征匹配过程,探索特征点位置分布质量的刻画与评价方法。【结果】在同一台工作站上运行Python 3.7和OpenCV 3.4.2.16程序,对设置相同参数的同一台无人机获取的阔叶和针叶林航摄影像,选择ORB、FAST、SURF、SIFT、KAZE、AKAZE搭配本研究提出的分区匹配策略与普通分级下的采样匹配策略进行对比试验,搭配分区匹配策略的6种特征算子相比普通策略平均匹配速度提升11.53%,平均匹配率提升0.83%,平均位置分布质量提升2.46%。【结论】低空无人机森林航摄影像与普通测绘航摄影像特点不同,不能套用普通测绘影像的匹配策略。采用原始分辨率影像分区匹配策略,在保留影像细节的同时可缩小匹配范围,提高匹配速度;借助无人机POS数据构建像对间单应约束,能够改善匹配特征点的位置分布质量、降低错误匹配率。搭配分区匹配策略的6种特征算子试验表明该策略有效、灵活、与特征算子无关,AKAZE算子搭配分区匹配策略效果最佳。
【Objective】In order to solve the problems of frequent occurrence of slow matching speed,high error matching rate and poor matching quality when processing aerial photographic images in forests of UAV in low altitude by directly using photogrammetric software,the article started from the aerial photographic image features in forests in low altitude to explore an image matching strategy suitable for UAV’s aerial photographic images in forests.【Method】The homography restraints among image pairs were built with the aid of UAV’s POS data based on the related principles of photogrammetry and computer vision.This is a partition matching strategy of which several corresponding subareas are divided according to certain location distribution rules and each sub area is regarded as an independent image to be processed during the feature extraction and matching processes.Meanwhile the GIS space analysis algorithm was introduced into the feature matching processes to explore the characterization and evaluation of location distribution quality of feature points.【Result】Under the software programming environments of the same work station Python 3.7 and OpenCV 3.4.2.16,the contrast experiment was implemented by using the six kinds of characteristic operators of ORB,FAST,SURF,SIFT,KAZE and AKAZE,with collaboration of the partition matching strategy and the sampling matching strategy under the common classification to the aerial photographic images of broad-leaved forest and coniferous forest with the same parameters set.The results showed that the average matching speed of the six kinds of operators in collaboration with partition matching strategy was increased by 11.53%,the average matching rate was increased by 0.83%,and the average location distribution quality was improved by 2.46%.【Conclusion】The low altitude UAV’s forest images presented different features from the common mapping aerial photographic images,where the matching methods of the common mapping images couldn’t be used discriminately.The par
作者
李亚东
曹明兰
李长青
冯仲科
Li Yadong;Cao Minglan;Li Changqing;Feng Zhongke(Beijing Key Laboratory of Precision Forestry,Beijing Forestry University,Beijing 100083;Beijing Polytechnic College,Beijing 100042)
出处
《林业科学》
EI
CAS
CSCD
北大核心
2020年第10期113-120,共8页
Scientia Silvae Sinicae
基金
中央高校基本科研业务费专项资金项目(2015ZCQ-LX-01)。
