改进的单目视觉实时定位与测图方法
发布时间：2018-10-15 13:06
【摘要】：针对经典单目实时定位与测图(SLAM)采用卡尔曼滤波(EKF)滤波和FAST特征角点所存在的非线性误差和鲁棒性较差的问题,提出了一种改进的单目视觉实时定位与测图方法。该方法采用相机中心的迭代EKF(IEKF)滤波方法,将特征点在当前相机坐标系下表达,并在线性化展开点附近迭代更新,不断逼近最优位置,从而最小化线性化误差;针对特征点跟踪的鲁棒性、高效性及分布不均的问题,选用具有尺度和旋转不变性,且探测和匹配效率更高的ORB特征作为特征角点,并采用一种由探测到筛选阶段的整体网格化处理方法;另外,采用特征点逆深度参数化方法,避免了因深度信息未知而导致的局部地图初始化错误问题,并采用1点随机抽样一致方法(RANSAC)滤波更新方法剔除错误的特征匹配,保证滤波估计的准确与稳定。实验采用外符合精度对算法进行评价,结果表明:新方法具有更强的鲁棒性,绝对定位精度提升至2.24 m,误差轨迹比提升至1.3%,且满足实时性要求,是一种实用性较强的单目视觉实时定位与测图方法。
[Abstract]:Aiming at the problem of poor nonlinear error and robustness of classical monocular real-time location and mapping (SLAM) using Kalman filter (EKF) filter and FAST feature corner, an improved monocular vision real-time location and mapping method is proposed. The method uses the camera center iterative EKF (IEKF) filtering method to express the feature points in the current camera coordinate system, and iteratively updates near the linearization expansion point to continuously approximate the optimal position, thus minimizing the linearization error. Aiming at the problem of robustness, high efficiency and uneven distribution of feature point tracking, ORB features with scale and rotation invariance and higher detection and matching efficiency are selected as feature corners. In addition, the method of inverse depth parameterization of feature points is used to avoid the problem of local map initialization error caused by unknown depth information. The one-point random sampling consistent method (RANSAC) filter updating method is used to eliminate the wrong feature matching to ensure the accuracy and stability of the filter estimation. The experimental results show that the new method is more robust, the absolute positioning accuracy is increased to 2.24 m, the error locus ratio is increased to 1.3%, and the real-time requirement is satisfied. It is a practical and practical method for real-time positioning and mapping of monocular vision.
【作者单位】： 信息工程大学导航与空天目标工程学院;
【基金】：国家自然科学基金(41274014,41501491)项目资助
【分类号】：TN713;TP391.41
，
本文编号：2272644
[Abstract]:Aiming at the problem of poor nonlinear error and robustness of classical monocular real-time location and mapping (SLAM) using Kalman filter (EKF) filter and FAST feature corner, an improved monocular vision real-time location and mapping method is proposed. The method uses the camera center iterative EKF (IEKF) filtering method to express the feature points in the current camera coordinate system, and iteratively updates near the linearization expansion point to continuously approximate the optimal position, thus minimizing the linearization error. Aiming at the problem of robustness, high efficiency and uneven distribution of feature point tracking, ORB features with scale and rotation invariance and higher detection and matching efficiency are selected as feature corners. In addition, the method of inverse depth parameterization of feature points is used to avoid the problem of local map initialization error caused by unknown depth information. The one-point random sampling consistent method (RANSAC) filter updating method is used to eliminate the wrong feature matching to ensure the accuracy and stability of the filter estimation. The experimental results show that the new method is more robust, the absolute positioning accuracy is increased to 2.24 m, the error locus ratio is increased to 1.3%, and the real-time requirement is satisfied. It is a practical and practical method for real-time positioning and mapping of monocular vision.
【作者单位】： 信息工程大学导航与空天目标工程学院;
【基金】：国家自然科学基金(41274014,41501491)项目资助
【分类号】：TN713;TP391.41
，
本文编号：2272644
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2272644.html
