【作者】 姜肖楠；

【导师】 黄显林；

【作者基本信息】 哈尔滨工业大学 ， 控制科学与工程， 2010， 博士

【摘要】 自主导航方法作为月球探测二、三期工程的核心技术,是月球探测器定点软着陆成败的关键。视觉导航方法经过近三十年的发展,在传感器技术、芯片计算能力显著提高的推动下,具备了低成本、低功耗、高可靠性的显著优点,已经被广泛的应用在地面机器人、无人机、水下机器人的导航系统中。随着视觉处理算法的日益成熟,将视觉导航系统应用于月球探测器软着陆任务中,能够大幅度的提高探测器定点软着陆的导航精度。本文对月球软着陆中的视觉导航相关问题及其仿真方法做出了较为系统的研究。针对月球软着陆的视觉导航和障碍识别仿真问题,提出了一种模拟月面生成方法。现有的月面地形信息无法应用在定点软着陆导航仿真中,算法将在现有的低分辨率月面高程数据的基础上,生成月面分形地形;之后,依据最新的月面地形数据库和地貌统计特征,添加月面的主要地形特征——撞击坑,并给出模拟月面的生成实例。在分析月球软着陆视觉导航任务需求的基础上,提出了采用改进SIFT特征算法的特征匹配初始定位方法。算法在高斯差分图像的blob区域中随机选取采样点,并通过有限次迭代逼近极值点,计算相应的描述子;通过对大量不同月面地形的图像特征提取和匹配实验确定并验证了算法在不同类型月面区域中的参数设定;之后,介绍了一种基于多匹配点初始定位问题的线性解法,并使用月球表面地形模型验证了特征匹配初始定位算法的可行性。提出了一种基于多传感器信息融合的月面障碍识别方法。在基于CCD成像的月面障碍识别研究中,主要提出了基于二维Renyi熵阈值分割的月面岩石识别方法和基于纹理分析的粗糙地形识别方法;同时在介绍一种激光成像雷达模型的基础上,提出了一种成像模拟方法,并给出了基于激光雷达数据的月面坡度计算方法,用以判断地形是否符合着陆要求;在此基础上,提出了一种基于模糊逻辑决策的CCD成像和激光雷达成像的信息融合算法,解决了多传感器条件下的月面地形安全性评估问题。最后,对基于月面特征匹配的探测器运动估计及其非线性滤波仿真实验方法进行了研究。所提出的视觉/惯性组合导航系统能够很好的估计探测器的位置和速度信息,但是其姿态估计效果不理想。为了加快导航系统姿态估计的收敛速度、提高组合导航系统的容错性、降低估计噪声,提出了一种基于联邦UKF滤波的视觉/惯性/天文组合导航方法,并通过仿真实验验证了联邦滤波器对导航系统姿态估计的性能改进。更多还原

【Abstract】 Autonomous navigation is one of the core technologies of the second and third phase of China Lunar Exploration Project, could determine the success of pin-point soft landing. With the development of three decades, driven by the sensor technology and significant improvement in chip’s computing power, visual navigation methods has got the advantages of low cost, low power and high reliability, which has been widely used in the navigation systems of ground robots, unmanned aerial vehicles and underwater autonomous vehicles. With the vision processing algorithms become more sophisticated, the visual navigation system used in lunar soft landing missions, can greatly improve the navigation accuracy of explorer’s soft pin-point landing. Thetelated problems of visual navigation and simulation methods will be researched. For the simulation of visual navigation and hazard detection in lunar soft landing, a modeling of terrain surface is proposed. The high-resolution digital elevation map will be generated from low-resolution data based on a square-square subdivision fractal function. And the lunar impact craters will be added to the model according to the latest database and the statistic results of observation. Some of the modeling results will be presented.Based on the analysis of soft landing visual navigation mission requirement, an improved SIFT feature detecting and matching algorithm is proposed. The samples are chosen in the "blob" region of Gaussian difference images. The extremum is obtained in limited times of iterative calculations. Then the relative descriptor is calculated. The parameter settings are established with large amount of features extraction and matching experiments on different types of lunar terrain surface. A linear pose estimation algorithm with multi matching features is introduced, which verified the feasibility of feature matching initialization algorithm on modeled lunar terrain surface.Then we presented a multi-sensor data fusion based lunar surface hazard detection methods. On the CCD imaging hazard detection, 2-D Renyi’s entropy threshold based rock detection and texture analysis based hazard detection methods are proposed. With an imaging LIDAR model introduced, we presented a LIDAR simulation methods and local surface gradient is calculated with LIDAR data to verify the safety of spacecraft’s landing. Then a fuzzy logic based fission decision algorithm for CCD and LIDAR information is developed, which solves the problem of lunar surface safety evaluation under multi-sensor condition.Finally, on the research of lunar surface feature matching motion estimation and relative nonlinear filters, we introduce a vision/inertial navigation system which estimated the position and velocity quite well, but the attitude estimation results are not ideal. To expedite the convergence rate and reduce the errors of attitude estimation, a federated UKF VNS/SINS/CNS filter is proposed. Simulations of federated filter illustrate the improvement of attitude estimation.更多还原