IEEE Trans Vis Comput Graph. 2018 Feb;24(2):1091-1102. doi: 10.1109/TVCG.2017.2657634. Epub 2017 Jan 24.
We propose a visual marker embedding method for the pose estimation of a projection surface to correctly map projected images onto the surface. Assuming that the surface is fabricated by a full-color or multi-material three-dimensional (3D) printer, we propose to automatically embed visual markers on the surface with mechanical accuracy. The appearance of the marker is designed such that the marker is detected by infrared cameras even when printed on a non-planar surface while its appearance can be diminished by the projection to be as imperceptible as possible to human observers. The marker placement is optimized using a genetic algorithm to maximize the number of valid viewpoints from which the pose of the object can be estimated correctly using a stereo camera system. We also propose a radiometric compensation technique to quickly diminish the marker appearance. Experimental results confirm that the pose of projection objects are correctly estimated while the appearance of the markers was diminished to an imperceptible level. At the same time, we confirmed the limitations of the current method; only one object can be handled, and pose estimation is not performed at interactive frame rates. Finally, we demonstrate the proposed technique to show that it works successfully for various surface shapes and target textures.
我们提出了一种用于投影表面姿态估计的视觉标记嵌入方法,以正确地将投影图像映射到表面上。假设表面是由全彩色或多材料三维(3D)打印机制造的,我们建议使用机械精度自动在表面上嵌入视觉标记。标记的外观设计使得即使在非平面表面上打印时也可以通过红外摄像机检测到标记,而其外观可以通过投影尽可能地减小,以使人类观察者难以察觉。使用遗传算法对标记位置进行优化,以最大限度地增加可以使用立体相机系统正确估计物体姿态的有效视点数量。我们还提出了一种辐射补偿技术,以快速减小标记的外观。实验结果证实,即使标记的外观减小到难以察觉的水平,投影物体的姿态也可以正确估计。同时,我们还确认了当前方法的局限性;只能处理一个物体,并且不能以交互帧率执行姿态估计。最后,我们演示了所提出的技术,以证明它可以成功地用于各种表面形状和目标纹理。