| 基于Kinect的改进移动机器人视觉SLAM |
| |
| 引用本文： | 蔡军,陈科宇,张毅.基于Kinect的改进移动机器人视觉SLAM[J].智能系统学报,2018,13(5):734-740. |
| |
| 作者姓名： | 蔡军 陈科宇 张毅 |
| |
| 作者单位： | 重庆邮电大学 重庆市信息无障碍与服务机器人工程技术研究中心, 重庆 400065 |
| |
| 摘 要： | 针对传统ICP（iterative closest points，迭代最近点算法）存在易陷入局部最优、匹配误差大等问题，提出了一种新的欧氏距离和角度阈值双重限制方法，并在此基础上构建了基于Kinect的室内移动机器人RGB-D SLAM（simultaneous localization and mapping）系统。首先，使用Kinect获取室内环境的彩色信息和深度信息，通过图像特征提取与匹配，结合相机内参与像素点深度值，建立三维点云对应关系；然后，利用RANSAC（random sample consensus）算法剔除外点，完成点云的初匹配；采用改进的点云配准算法完成点云的精匹配；最后，在关键帧选取中引入权重，结合g2o（general graph optimization）算法对机器人位姿进行优化。实验证明该方法的有效性与可行性，提高了三维点云地图的精度，并估计出了机器人运行轨迹。
|
| 关 键 词： | 移动机器人 Kinect 同时定位与地图构建 迭代最近点算法 关键帧 随机采样一致性 位姿估计 三维重建 |
Improved V-SLAM for mobile robots based on Kinect |
| |
| CAI Jun,CHEN Keyu,ZHANG Yi.Improved V-SLAM for mobile robots based on Kinect[J].CAAL Transactions on Intelligent Systems,2018,13(5):734-740. |
| |
| Authors: | CAI Jun CHEN Keyu ZHANG Yi |
| |
| Affiliation: | Chongqing Information Accessibility and Service Robot Engineering Technology Research Center, Chongqing University of Posts and Telecommunications, Chongqing 400065, China |
| |
| Abstract: | Given that the traditional iterative closest points (ICP) algorithm easily falls into the local optimum and has a large matching error, a novel double restriction method containing Euclidean distance and angle threshold is proposed. To realize this, an indoor mobile robot RGB-D SLAM (simultaneous localization and mapping) using a Kinect camera was developed. First, the Kinect camera was used to get color information and depth information for the indoor environment. Through the image feature extraction and matching procedure, the relationship between two 3D point clouds was established by combining the camera intrinsic parameters and pixel depth values. Then, the initial registration was completed using the random sample consensus (RANSAC) algorithm to remove outliers. Meanwhile, accurate registration was completed using the improved ICP algorithm. Finally, the weight was introduced into the selection of the key frames, and the general graph optimization (g2o) algorithm was used to optimize the pose of the robot. The experimental results prove effectiveness and feasibility of the method, and this method improves the accuracy of the 3D point cloud map and estimates the trajectory of the robot. |
| |
| Keywords: | mobile robot Kinect SLAM ICP key-frame RANSAC pose estimate three-dimensional reconstruction |
|
| 点击此处可从《智能系统学报》浏览原始摘要信息 |
|
点击此处可从《智能系统学报》下载全文 |
