SIA OpenIR  > 水下机器人研究室
基于前视声纳的水下环境地图构建方法研究
Alternative TitleResearch on Underwater Environmental Map Construction Using Forward-looking Sonar
蒋敏
Department水下机器人研究室
Thesis Advisor封锡盛 ; 李一平
Keyword水下机器人 前视声纳 声纳图像配准 声学闭环检测 声光双目
Pages116页
Degree Discipline机械电子工程
Degree Name博士
2020-05-29
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract广袤的海洋覆盖了地球表面积的71%,一直以来人们从未停止对未知海洋探索的脚步。作为人类了解和认识海洋的一种基本工具,高质量水下环境地图可以满足水下考古、海洋科学研究及海洋军事等不同领域的需求。一方面,水下机器人具有近底航行的能力,在其上搭载成像传感器构建的水下环境地图比利用其它载体构建的地图质量更高。另一方面,水下机器人利用同步定位与地图构建(Simultaneous Localization and Mapping, SLAM)技术可以在不依赖于诸如声学信标或者水面船等外部辅助的情况下进一步提高地图构建的精度。在给定环境观测值和里程计信息的情况下,就如何估计机器人状态、环境地图的联合概率分布而言,水下SLAM和陆地SLAM没有本质的区别。但是,相较于陆地上常用的激光雷达、相机等高精度的传感器来说,水下声纳传感器生成的图像分辨率较低、噪声较多而且水下环境多为非结构化自然环境。这就造成陆地SLAM中一些成熟的图像配准算法和闭环检测方法不能够直接应用到水下环境中。针对这些问题,本文开展了如下三方面内容的研究。(1)基于对称KL散度的机械扫描声纳点云配准。机械扫描声纳具有结构紧凑、能量消耗低的优点,从而被广泛应用于便携型、经济型以及长航程型水下机器人,用于观测浑浊的水下环境。但是,机械扫描声纳所成图像在空间和时间维度上精度不高,导致机械扫描声纳图像的配准比较困难。本文提出一种名为SKLD-D2D的机械扫描声纳点云配准算法,它将声纳点云建模为高斯混合模型,以概率分布到概率分布的模式配准两帧声纳点云。采用对称KL散度来衡量两个高斯混合模型之间的距离,对称KL散度不仅具有KL散度适用于衡量两个概率分布间距离的优点同时又克服了KL散度的不对称性缺点。另外,本文利用了一种近似策略来获得两个高斯混合模型之间对称KL散度的闭式解。在采集自真实环境声纳图像数据集上证明了SKLD-D2D配准算法可以在保证算法精度的情况下大大降低算法的计算代价。(2)基于PHD滤波器的声学闭环检测及水下地图构建。目前,水下SLAM研究领域常常依赖于船位推算或惯性导航估计的机器人位姿来检测闭环。但是,机器人位姿估计误差本身又会随时间而无限制地增加。本文提出一种不依赖于机器人位姿估计而采用机械扫描声纳图像数据关联来进行闭环检测的方法。首先,将当前声纳图像与历史所有声纳图像进行相似性得分计算构建相似性矩阵以呈现闭环图像序列。为了能够评价非结构化图像的相似性,将两幅声纳图像之间的粗配准误差定义为它们之间的相似性得分。在粗配准步骤中,本文提出了两个新颖的特征,即强度投影直方图和极性梯度矩阵,结合互相关技术可分别计算平移参数和旋转参数。其次,机械扫描声纳的扫描距离较远,相似性矩阵中可能会同时呈现出多条闭环图像序列。本文提出采用概率假设滤波器(Probability Hypothesis Density, PHD)从相似性矩阵中提取这些闭环图像序列。再次,本文设计了三条启发式策略用以筛选上一步得到的闭环约束并将经过验证的闭环约束加入到GraphSLAM框架中通过约束优化调整每幅声纳图像对应的水下机器人位姿,从而获得全局一致水下地图。最后,本文在采集自结构化和非结构化环境的数据集上证明了所提闭环检测方法的有效性。(3)面向三维声学地图构建的声光信息融合点云提取。要将基于二维前视声纳的水下环境地图构建方法推广到水下声学三维地图构建,只依赖于前视声纳是较难的。这是因为,声纳传感器在成像过程中丢失了成像目标高度信息。考虑到水下机器人往往会同时搭载光学相机,而相机在成像过程中丢失了深度信息。将声纳和相机安装为双目形式可以充分利用它们之间的互补信息恢复出成像目标的三维点云信息,在此基础上可以将二维地图构建方法向三维声学地图构建扩展。首先,估计声纳和相机之间的外标定参数。现有标定方法优化目标函数存在局部极小值附近较为平坦的现象,使得目标函数对声纳坐标系下标定平面法向量的模长变化不敏感而出现尺度因子模糊问题,进一步影响声纳图像特征点高度角估计,导致外标定参数估计不准确。针对该问题,本文从声纳图像中选择一对特征点来解决声纳坐标系下的尺度因子模糊问题。其次,还推导了声纳图像和相机图像之间的对极几何约束关系。最后,在声光匹配特征点基础上,采用距离解法恢复成像目标三维坐标点。仿真实验验证了本文方法的有效性。
Other AbstractAbout 71 percentage of the Earth's surface is covered by the ocean. And it has been a very long time since our ancient ancestors set their sights to the vast ocean and desired to disclose the mystery underneath the water surface. As a tool for learning about the ocean, high-quality underwater environmental map meets demand from various fields such as underwater archaeology, marine scientific research, and marine military. On the one hand, the underwater environmental map constructed by the unmanned underwater vehicle equipped with the mapping sensor is of higher quality than the maps generated by the other methods due to its capability of operating in close proximity to the seafloor, on the other hand, the accuracy of the map can be further improved with the simultaneous localization and mapping(SLAM) technique independent of the external aids such as the acoustic beacon and supporting vessel. Given the environmental observation and odometry observation, there is no difference to the estimation of the joint probability of the vehicle state and environmental map between the underwater SLAM and terrestrial SLAM. However, the underwater sonar image has lower resolution and more noise compared to the data generated by high resolution sensors, i.e., laser sensor and optical camera used by terrestrial robot. In addition, underwater environment is mostly unstructured natural environment. As a result, the mature image registration algorithms and loop-closure detection methods used in terrestrial SLAM cannot be directly applied to underwater SLAM. To solve these problems, this paper has carried out the following three aspects of research. (1) Scan registration for underwater mechanical scanning imaging sonar using symmetrical Kullback–Leibler divergence Due to its advantages in size and energy consumption, mechanical scanning imaging sonar (MSIS) has been widely used in portable and economic unmanned underwater vehicles to observe the turbid and noisy underwater environment. However, handicapped by the coarseness in spatial and temporal resolution, it is difficult to stitch the scan pieces together into a panoramic map for global understanding. A registration method named symmetrical Kullback–Leibler divergence (SKLD)-distribution-to-distribution (D2D), which models each scan as a Gaussian mixture model (GMM) and evaluates the similarity between two GMMs in a D2D way with the measure defined by SKLD, is proposed to register the scans collected by MSIS. SKLD not only has the advantage that KL divergence is suitable for measuring the distance between two probability distributions, but also overcomes the disadvantage of asymmetry of KL divergence. Moreover, an approximation strategy is designed to derive a tractable solution for the KLD between two GMMs. Experimental results on the scans that were collected from the realistic underwater environment demonstrate that SKLD-D2D dramatically reduces the computational cost without compromising the estimation precision. (2) Underwater loop-closure detection for mechanical scanning imaging sonar by filtering the similarity matrix with probability hypothesis density filter. Most underwater SLAM implementations rely on the vehicle pose information estimated by dead reckoning or inertial navigation system to detect loop-closure. However, with time goes by, the pose estimation error accumulated unboundedly. This paper proposed a method using sonar image association to detect loop-closure, independent of the vehicle pose estimation. Firstly, a similarity matrix for pairs of images is constructed to represent the loop-closing tracks. The similarity score of two images is defined as their coarse registration error. In the registration step, two novel features, namely the intensity projection histograms and a polar gradient matrix, are extracted to calculate the translational and rotational parameters respectively. Secondly, the probability hypothesis density (PHD) filter is used to extract the possible loop-closure constraints from the similarity matrix, removing the random noise brought by accidental correlation and refining the concurrent loop-closing tracks resulted from long-range scanning. Thirdly, three heuristic strategies are designed to validate the loop-closure candidates. Lastly, the loop-closure constraints from validated candidates are fed into the GraphSLAM system to adjust the pose of each scan by constraint optimization, then, self-consistency underwater map can be obtained. Experiments on the MSIS sonar images collected in structured and unstructured underwater environments validate the effectiveness of the proposed loop-closure detection method. (3) Point cloud extraction by opti-acoustic information fusion for 3D acoustic map construction. It is difficult to extend the underwater 2D map construction method to the 3D map construction using only forward-looking sonar. Because the target height information is lost during the sonar image formation. On the other hand, unmanned underwater vehicles are often equipped with optical camera and depth information is lost during the optical image formation. Installing the sonar and camera in binocular form can extract the 3D point cloud of the imaging target from the sonar and camera image. The 3D acoustic point cloud is beneficial to construct 3D acoustic map. Firstly, the external calibration parameters of the opti-acoustic stereo need to be estimated. The objective function value of existing method doesn't change much near the local minimum, making the objective function insensitive to the change of the magnitude of the calibration plane's normal vector. Therefore, the elevation angle of the selected corresponding feature point of the sonar/camera image cannot be estimated accurately, leading to inaccurate estimation of external calibration parameters. To solve this problem, this paper selects a pair of feature points from the sonar image to correctly estimate the magnitude of the normal vector. Secondly, the constraint equations for the epipolar geometry of opti-acoustic stereo are derived. Lastly, the three dimensional coordinates of the feature point are recovered by the "distance solution" based on the opti-acoustic feature correspondences. Simulation experiments verified the effectiveness of the proposed method.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/27166
Collection水下机器人研究室
Affiliation中国科学院沈阳自动化研究所
Recommended Citation
GB/T 7714
蒋敏. 基于前视声纳的水下环境地图构建方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2020.
Files in This Item:
File Name/Size DocType Version Access License
基于前视声纳的水下环境地图构建方法研究.(20774KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[蒋敏]'s Articles
Baidu academic
Similar articles in Baidu academic
[蒋敏]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[蒋敏]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.