基于双目强约束的直接稀疏视觉里程计

为了提高双目直接稀疏里程计(Stereo Direct Sparse Odometry, Stereo DSO)的定位速度和精度, 使得移动机器人可以更有效地执行任务, 提出了一种基于双目强约束的直接稀疏视觉里程计系统。基于直接法的即时定位与地图构建(Simultaneous Localization and Mapping, SLAM)系统直接对图像像素构建光度误差优化函数, 无需提取特征点, 克服了基于特征点法的SLAM系统在弱纹理场景下不鲁棒的缺陷, 并且在前端跟踪阶段效率更高。提出一种快速、准确的双目初始化方法, 结合三角化不确定性为不同类型的点赋予不同的深度范围, 加速深度滤波器的收敛。同时, 在运动估计阶段引入双目约束, 使得该系统在绝对尺度上的定位更加准确。通过在公开的KITTI数据集11个序列上进行实验, 实验结果表明所提出的算法在定位精度上明显优于同样采用直接法的Stereo Large Scale Direct SLAM(LSD-SLAM2)和Stereo DSO, 并达到与基于特征点法的ORB-SLAM3相近的水平, 为直接法SLAM提供一种更优的定位方案。     

基于分级关键帧筛选的RGB-DSLAM算法研究

利用关键帧求解SLAM算法(simultaneous localization and mapping)能够提高SLAM系统的实时性与精确度。针对现存关键帧筛选算法中存在的计算复杂度高、图像帧冗余以及鲁棒性较差等问题,提出一种分级关键帧筛选方法。该算法考虑了SLAM系统在不同运行阶段时对关键帧的要求,首先结合旋转度指数与地图点跟踪筛选出一级关键帧用于后端优化与回环检测,再利用相邻帧在空间上的相对运动距离筛选出二级关键帧用于三维地图构建,最后,实现了基于此二级筛选算法的RGB-D SLAM系统。实验表明,一级关键帧算法能提高SLAM系统的定位和建图精度,二级关键帧算法则有效减少了数据冗余,提高了建图效率。     

基于噪声缩放的自适应UKF-SLAM算法

针对扩展卡尔曼滤波(EKF)算法在移动机器人同时定位和环境建模(SLAM)中的缺点,即非线性系统简单线性化所导致的系统状态方程的不准确性、雅克比矩阵的计算所导致的计算复杂化以及噪声模型不确定性所导致的滤波稳定性降低等问题,提出一种对噪声自适应的UKF-SLAM算法。该算法通过对噪声缩放进而改变噪声模型,利用观测残差序列准确估计观测噪声模型协方差,运用预测的新息协方差和IAE开窗法求其系统状态噪声缩放因子,从而准确估计系统状态噪声模型协方差,实现对不确定的噪声模型能够自适应UKF-SLAM算法。UKF的Sigma点采样策略是比例对称采样。实验结果证明,该方法相对EKF算法和UKF算法具有较高的定位精度和自适应能力。     

一种融合多帧ICP和图优化的算法研究

当前基于迭代最近点拼接的同时定位与建图算法,存在误差积累、无法满足大范围定位精度的缺陷。为此,提出一种融合多帧迭代最近点和图优化的算法。在时域上处理点云拼接问题,将单帧迭代最近点算法推广到多帧进行最近点迭代,提取同一地点在不同时刻的数据特征,形成多个封闭循环,再运用基于最小二乘的图优化方法对点云拼接后的全网数据进行全局优化,消除累计误差,提升整体的定位精度。采用鲁巷和密歇根的数据进行测试,结果表明,该方法在一定程度上减少了匹配误差,平均误差为1.0m,最小误差为0.2m,可以满足大范围同步定位与建图的精度需求。     

一种非迭代的位姿估计方法

针对深度相机运动中的位姿估计问题,提出了一种无需迭代的估计方法。首先,在二维图像上应用图像特征点提取和描述方法,完成不同视点的初始匹配。其次,选择初始匹配度量距离最小的2个特征点作为种子点。以三维空间中欧式距离与坐标系的建立无关为准则,对初始匹配进行筛选。剔除误匹配点对,进而计算运动位姿参数。最后,采用nyuv2图像数据库进行实验,验证了本文算法的可行性和正确性。实验结果表明:与传统算法相比,该方法计算效率平均提高了8倍以上,特别适用于大型场景中的同步定位和地图构建SLAM(Simultaneous Localization and Mapping)问题。     

SLAM无损检测技术的研究

本文分析了声波对有损耗样品板的透射规律，得到了透射率|W|^2的公式。对于两类样品板，通过数值计算得到了它们的响应曲线|W|^2θ，通过调节入射角θ1得到|W|^2的最大值。所做工作有利于提高SLAM的检测灵敏度。     

Semi-Direct Visual Odometry and Mapping System with RGB-D Camera

In this paper a semi-direct visual odometry and mapping system is proposed with a RGB-D camera, which combines the merits of both feature based and direct based methods. The presented system directly estimates the camera motion of two consecutive RGB-D frames by minimizing the photometric error. To permit outliers and noise, a robust sensor model built upon the t-distribution and an error function mixing depth and photometric errors are used to enhance the accuracy and robustness. Local graph optimization based on key frames is used to reduce the accumulative error and refine the local map. The loop closure detection method, which combines the appearance similarity method and spatial location constraints method, increases the speed of detection. Experimental results demonstrate that the proposed approach achieves higher accuracy on the motion estimation and environment reconstruction compared to the other state-of-the-art methods. Moreover, the proposed approach works in real-time on a laptop without a GPU, which makes it attractive for robots equipped with limited computational resources.     

基于即时稠密三维重构的无人机视觉定位

传统景象匹配定位方法在用于低空无人机定位时,易因低空航拍图像视场小,且与卫星图像（带有地理信息）的拍摄角度差异大而失败.本文提出了一种基于即时稠密三维重构的无人机视觉定位方法,通过将稠密三维点云与卫星图像匹配以实现无人机定位.首先根据图像序列快速估计摄像机位姿,而后使用多深度图协同去噪与优化算法生成稠密三维点云,随后通过变换观察视角由稠密三维点云生成与卫星图像拍摄视角相近的虚拟视图,最后将虚拟视图与卫星图像匹配并得到无人机的地理坐标.由于稠密三维点云包含多张图像的信息,覆盖面积大,且可变化观察视角,因此能够有效克服上述两个问题.实验证明了本文方法的有效性.     

Large scale multiple robot visual mapping with heterogeneous landmarks in semi-structured terrain

This paper addresses the cooperative localization and visual mapping problem with multiple heterogeneous robots. The approach is designed to deal with the challenging large semi-structured outdoors environments in which aerial/ground ensembles are to evolve. We propose the use of heterogeneous visual landmarks, points and line segments, to achieve effective cooperation in such environments. A large-scale SLAM algorithm is generalized to handle multiple robots, in which a global graph maintains the relative relationships between a series of local sub-maps built by the different robots. The key issue when dealing with multiple robots is to find the link between them, and to integrate these relations to maintain the overall geometric consistency; the events that introduce these links on the global graph are described in detail. Monocular cameras are considered as the primary extereoceptive sensor. In order to achieve the undelayed initialization required by the bearing-only observations, the well-known inverse-depth parametrization is adopted to estimate 3D points. Similarly, to estimate 3D line segments, we present a novel parametrization based on anchored Plücker coordinates, to which extensible endpoints are added. Extensive simulations show the proposed developments, and the overall approach is illustrated using real-data taken with a helicopter and a ground rover.     

Amortized constant time state estimation in Pose SLAM and hierarchical SLAM using a mixed Kalman-information filter

The computational bottleneck in all information-based algorithms for simultaneous localization and mapping (SLAM) is the recovery of the state mean and covariance. The mean is needed to evaluate model Jacobians and the covariance is needed to generate data association hypotheses. In general, recovering the state mean and covariance requires the inversion of a matrix with the size of the state, which is computationally too expensive in time and memory for large problems. Exactly sparse state representations, such as that of Pose SLAM, alleviate the cost of state recovery either in time or in memory, but not in both. In this paper, we present an approach to state estimation that is linear both in execution time and in memory footprint at loop closure, and constant otherwise. The method relies on a state representation that combines the Kalman and the information-based approaches. The strategy is valid for any SLAM system that maintains constraints between marginal states at different time slices. This includes both Pose SLAM, the variant of SLAM where only the robot trajectory is estimated, and hierarchical techniques in which submaps are registered with a network of relative geometric constraints.