论文部分内容阅读
提出了一种基于RGB-D数据的实时SLAM(同时定位与地图创建)算法,得到机器人的6D位姿并构建环境的3D地图.首先提取RGB图像的特征点,并利用高斯混合模型得到特征点处的颜色和3维坐标,以及对应的协方差矩阵.然后利用ICP(迭代最近点)算法,得到当前帧特征点与环境模型特征点集的变换矩阵T_t,并在T_t周围撒点采样,得到观测最优的传感器位姿矩阵P_t.之后利用P_t将当前帧的密集点云变换到全局坐标系下,构建环境3D地图.最后,利用卡尔曼滤波器对环境模型特征点集进行更新.为防止特征点较少时,ICP算法误差较大,本文加入粒子采样步骤,有效地提高了定位精度.此外,在将当前帧特征点与环境模型特征点进行数据关联时,引入颜色信息提高了数据关联的准确率.针对FR1基准包,本文算法的最小定位误差为1.7 cm,平均定位误差为11.9 cm,每帧数据平均处理时间为31 ms,可以满足机器人实时SLAM的要求.
A real-time SLAM algorithm based on RGB-D data was proposed to get the 6D pose of the robot and construct a 3D map of the environment.First, the feature points of the RGB image were extracted and the feature points were obtained by the Gaussian mixture model And the corresponding covariance matrix.And then using the ICP (iterative nearest-neighbor) algorithm, the transformation matrix T_t of the feature points of the current frame and the environment model feature point set is obtained, and the sampling point is sampled around T_t to obtain The optimal sensor pose and attitude matrix P_t is observed, and then the dense point cloud of the current frame is transformed into the global coordinate system by P_t to construct the 3D map of the environment.Finally, the set of environmental model feature points is updated by using Kalman filter, When the number of feature points is small, the error of ICP algorithm is larger, and the particle sampling step is added in this paper, which effectively improves the positioning accuracy.In addition, when the current frame feature points and the environmental model feature points are data-related, the introduction of color information improves the data association For the FR1 benchmark package, the proposed algorithm has a minimum positioning error of 1.7 cm, an average positioning error of 11.9 cm and an average data processing time of 31 ms per frame, People Live SLAM requirements.