Improving the accuracy of visual markers by four dots and image interpolation
Improving the accuracy of visual markers by four dots and image interpolation
复制标题
DOI：
10.1109/iris.2016.8066087
复制
发表时间：
2016-12
期刊：
影响因子：
--
通讯作者：
Hideyuki Tanaka;Kunihiro Ogata;Y. Matsumoto
中科院分区：
文献类型：
--
作者：
Hideyuki Tanaka;Kunihiro Ogata;Y. Matsumoto
视觉标记是支持自主机器人进行物体识别和姿势测量的有用工具；然而，传统标记在姿势估计方面的精度相对较低。通过改变标记器的设计和相应的图像处理算法，提高了估计精度。更具体地说，我们使用了四个点，这比使用给定标记的四个角更容易检测到。此外，我们使用了一种图像内插技术，以亚像素精度检测这些点的位置。结果表明，位姿估计误差约为传统标记器的10%。此外，我们观察到距离和光照变化方面的稳健性也得到了增强。通过添加四个点并结合我们的图像处理例程，我们的方法很容易适用于传统的标记系统。
Visual markers are useful tools that support object recognition and pose measurements by autonomous robots; however, conventional markers have relatively low levels of accuracy in terms of pose estimation. We improved this estimation accuracy by changing the design of the marker and the corresponding image processing algorithm. More specifically, we used four dots, which are more stably detectable than using four corners of a given marker. Further, we used an image interpolation technique to detect the positions of these dots with sub-pixel accuracy. As a result, pose estimation error reduced to approximately 10 percent of the error of conventional markers. Further, we observed that robustness in terms of distance and illumination variation was also enhanced. Our method is easily applicable to conventional marker systems by adding the four dots and incorporating our image processing routines.