Fiducial markers (AR/visual markers) are still useful tools in many AR/MR applications. But conventional markers have two fundamental problems in orientation estimation. One is degradation of orientation accuracy in frontal observation. The other is ``pose ambiguity" where the estimated orientation repeats switching between two values. We previously developed a novel marker ``ArrayMark" which uses a microlens array and solved the former problem. This time we propose a practical solution to the latter problem by improving the ArrayMark. We attach an additional reference point to detect the occurrence of invalid estimation, and modify the orientation of the marker by inverting the zenith-angle of the visual-line. This marker enables stable pose estimation from a one-shot image without using any filtering techniques. The method is applicable to conventional markers, too. We demonstrated the availability of this improvement to the pose-ambiguity problem.