APP下载

一种基于三维视觉的移动机器人定位与建图方法

2020-08-03陈超张伟伟徐军

现代电子技术 2020年6期
关键词:移动机器人

陈超 张伟伟 徐军

摘  要: 针对移动机器人三维视觉SLAM(同步定位与建图)中定位精度低、实时性差等问题,提出一种基于由初到精的位姿估计和双重闭环策略的SLAM方法。首先对MSER(最大稳定极值区)检测算法进行椭圆拟合化处理并提取出图像中的ROI(感兴趣区);然后从ROI中提取出稀疏像素点并使用直接法得到初始位姿变换参数;接着结合改进的基于八叉树结构的ICP(迭代最近点)对相机位姿进行精估计;再结合关键帧选择机制提出一种双重闭环检测方法为构建的位姿图添加约束;最后通过g2o图优化框架对位姿图进行优化并完成点云的拼接。通过NYU和TUM标准数据集验证了算法的实时性与有效性,室内实验结果表明,在复杂环境下也能利用该方法进行准确的位姿估计,并构建出环境的三维点云地图。

关键词: SLAM; 移动机器人; 三维视觉; 位姿估计; 闭环检测; 点云拼接

中图分类号: TN95?34; TP242                    文献标识码: A                      文章编号: 1004?373X(2020)06?0034?05

A method of mobile robot localization and mapping based on 3D vision

CHEN Chao, ZHANG Weiwei, XU Jun

(School of Mechanical Engineering, Jiangsu University of Science and Technology, Zhenjiang 212000, China)

Abstract: A SLAM method based on the initial to precise pose estimation and double closed?loop strategy is proposed to improve the positioning accuracy and real?time performance in the 3D vision SLAM (synchronous localization and mapping) of mobile robot. The ellipse fitting processing is performed for the MSER (maximally stable extremal region) detection algorithm and the ROI (region of interest) in the image is extracted, and then the sparse pixels are extracted from ROI and the initial pose transformation parameters are obtained by means of the direct method. The camera pose is accurately estimated in combination with the improved ICP (iterative nearest point) based on octree structure. A double closed?loop detection method is proposed in combination with the key frame selection mechanism to add constraints to the constructed pose map. By means of the g2o map optimization framework, the pose map is optimized and the splicing of point cloud is completed. The real?time performance and effectiveness of the algorithm is verified with the NYU and TUM standard data sets. The results of indoor experiments show that in a complex environment, this method can also be used to accurately estimate the pose and construct a 3D point cloud map of the environment.

Keywords: SLAM; mobile robot; 3D vision; pose estimation; closed?loop detection; point cloud splicing

0  引  言

SLAM(同步定位与建图)主要指机器人在未知環境中,通过携带的传感器获取外界信息以完成对自身位姿的估计,并构建出周围环境的地图[1]。其中,以相机进行信息采集的方法称为视觉SLAM[2],该方法可实现在复杂环境下的三维地图构建。

视觉SLAM大体分为两种,即特征点法与直接法。Artal 等人开源的ORB?SLAM2算法采用多线程处理[3],虽能构建相对稠密的地图,但算法的运行效率较低。相较之下,直接法通过对像素进行操作,能较好地应对特征点缺失的情况。文献[4]提出一种稀疏直接SLAM算法DSO,该算法通过优化相机内参、位姿和地图几何参数,可完成定位并构建半稠密的地图,但算法运行于单目相机且实时性较差。……

登录APP查看全文

猜你喜欢

移动机器人
移动机器人自主动态避障方法
移动机器人VSLAM和VISLAM技术综述
基于改进强化学习的移动机器人路径规划方法
基于ROS与深度学习的移动机器人目标识别系统
基于Twincat的移动机器人制孔系统
室内环境下移动机器人三维视觉SLAM
简述轮式移动机器人控制系统中的传感器
增强D*Lite在自主移动机器人安全路径规划中应用
室外移动机器人自主充电系统研究