论文部分内容阅读
面向服务机器人使用廉价的RGB-D摄像头自主构建室内3维地图的需求,提出一种鲁棒的图像对齐方法.基于特征点的匹配集计算帧间变换,使用随机抽样一致算法(RANSAC)消除误配,并改变其内点计数策略以适应特征点空间分布不均;同时检测地面信息,利用共面约束来增强点集对齐.在机器人从真实室内环境中采集的RGB-D图像序列上进行了实验,帧间对齐错误率为0,全局地面误差不超过2 cm;3维建图过程准确且能够连续进行.结果表明使用地面信息能有效提高地图的全局精度,方法兼备鲁棒性和准确性.
Aiming at the demand of service robots to build an indoor 3-D map using inexpensive RGB-D camera, a robust image alignment method is proposed.Firstly, based on matching points of feature points, inter-frame transform is calculated, and RANSAC is used to eliminate errors And change the counting strategy of its inner points to adapt to the spatial distribution of feature points unevenly.At the same time detect the ground information and use the coplanar constraint to enhance the point set alignment.In the RGB-D image sequences collected by the robot from the real indoor environment The error between the experiment and the inter-frame alignment is 0, and the global ground error is less than 2 cm. The 3-D construction process is accurate and continuous. The results show that using the ground information can effectively improve the global accuracy of the map. The method is robust and accurate .