摘要
基于特征点的视觉里程计在点特征稀少的环境下难以得到足够的匹配点对,从而导致相机运动估计失败,因而提出采集人造环境中特征明显的边缘作为线特征来提高视觉里程计算法的稳定性。采用深度相机获取的RGB图像进行LSD线特征提取,推断线特征对应的图像位置的深度信息,避免深度缺失,将线段上的2D点反投影为3D点,拟合3D点为三维直线,利用线特征匹配关系进行位姿估计。此外在位姿优化部分进行改进,利用拟合直线过程中的最佳过点,以及重投影的线段与观测线段的角度误差信息,推导了误差关于位姿扰动的雅克比矩阵,在图优化时利用重投影误差优化相机位姿,拓展了传统的优化方法。基于TUM缺少点特征的数据集的实验结果表明所提出的线特征视觉里程计方法相比ORB-SLAM2的轨迹估计精度提高63%,并能完整地跟踪轨迹。实验结果表明所提出算法在欠特征点环境中表现出了较高的精度和稳定性。
It is difficult to get enough matching points in the environment of sparse feature points,which leads to the failure of camera motion estimation.However,the stability of the algorithm can be improved by collecting the obvious edge features in the artificial environment as line features.The RGB image obtained by depth camera is used for LSD line feature extraction,and the depth information of the image position corresponding to the line feature is inferred to avoid depth missing.The 2D points on the line segment are back projected into 3D points,and the 3D points are fitted into 3D lines.The pose estimation is performed by using the line feature matching relationship.In addition,the position and pose optimization part is improved.By using the best crossing point in the process of fitting the line and the angle error information between the re-projected line segment and the observed line segment,the Jacobian matrix of the error about the position and pose disturbance is derived.In the graph optimization,the position and pose of the camera is optimized by using the reprojection error,which expands the traditional optimization method.Experimental results on the TUM dataset with less point feature show that the trajectory estimation accuracy of the proposed method is 63%higher than that of ORB-SLAM2,and the trajectory can be completely tracked.The experimental results show that the algorithm has high accuracy and stability in the environment of less feature points.
作者
黄平
曹镇
黄俊杰
HUANG Ping;CAO Zhen;HUANG Junjie(Harbin Engineering University College of Intelligent Systems Science and Engineering,Harbin 150001,China)
出处
《中国惯性技术学报》
EI
CSCD
北大核心
2021年第3期340-349,共10页
Journal of Chinese Inertial Technology
基金
国家自然科学基金(61871143)
黑龙江省自然科学基金(LH2019F006)
哈尔滨市应用技术研究与开发项目(2017R-AQXJ095)。
关键词
视觉里程计
深度相机
线特征
图优化
visual odometry
depth camera
line character
graph-based optimization
