摘要
基于光学图像的地形匹配导航方法是目前深空探测任务中实现精确着陆的一种主要方法,通过天体表面光学图像得到的陆标能够获取完备的探测器位置和姿态信息。因此,天体表面陆标提取技术是实现精确着陆的关键技术。为此,提出了一种陆标自动提取方法。首先,利用AKAZE特征检测子对天体表面图像进行多尺度特征点检测,该特征点对图像尺度变化、旋转都具有一定的鲁棒性;然后,引入密度聚类理论,结合DBSCAN聚类算法对AKAZE特征点进行聚类得到多个不同的聚类族;最后,通过设定阈值确定候选陆标和最终陆标,并提出了一种陆标提取性能评价方法。试验结果表明,该方法能够提取出具有较好的抗旋转性能和抗尺度变化性能的陆标特征,能为火星着陆过程中光学敏感器可能出现的多尺度、多视点变化带来的匹配鲁棒性差的问题提供相应的解决方案,对深空探测过程中实现精确着陆具有一定的参考意义。
The terrain matching navigation method based on optical image is one of the main methods to realize accurate landing in deep space exploration task. The landmarks obtained by the celestial surface optical image can produce complete detector position and attitude information. Therefore, the extraction technology of the celestial surface landmark is the key technology to achieve accurate landing. To this end, a robust automatic method to extract roadmap is proposed in the paper. Firstly, the multi-scale feature points of the celestial surface image are detected using the AKAZE feature detector. The feature points have some robustness to the image distortion and rotation. Then, by introducing the density clustering theory and combining with the DBSCAN clustering algorithm, the AKAZE feature points are clustered to get a number of different clusters. Finally, the candidate signs and final landmarks are determined by setting thresholds. And then an evaluation method of landmark extraction performance is proposed. The experimental results show that this method have good anti-rotation and anti-scale performance in extracting the landmark characteristics, and can provide strong technical support for the multi-scale and multi-view changes that the optical sensor may appear during the landing on Mars, and it is of theoretical significance to realize the accurate landing in deep space exploration.
作者
陶江
曹云峰
丁萌
庄丽葵
张洲宇
钟佩仪
TAO Jiang CAO Yunfeng DING Meng ZHUANG Likui ZHANG Zhouyu ZHONG Peiyi(Nanjing University of Aeronautics and Astronautics, Nanjing 210016, Chin)
出处
《航天返回与遥感》
CSCD
2018年第1期112-120,共9页
Spacecraft Recovery & Remote Sensing
基金
国家自然科学基金(No.61673211)
中国运载火箭技术研究院高校联合创新基金
关键词
特征检测
密度聚类
陆标提取
精确着陆
深空探测
feature detection
density clustering
landmark extraction
accurate landing
deep space explore
