This paper proposes a series of techniques for improving the precision of optical fiducial tracking on tangible tabletops. The motivation is to enable convincing interactive projection mapping on tangibles on the table, which requires a high precision tracking of the location of tangibles. We propose a new fiducial design optimized for GPU based tracking, a technique for calibrating light that allows for computation on a greyscale image rather than a binarized black and white image, an automated technique for compensating for optical distortions in the camera lenses, and a tracking algorithm implemented primarily in shaders on the GPU. The techniques are realized in the BullsEye computer vision software. We demonstrate experimentally that BullsEye provides sub-pixel accuracy down to a tenth of a pixel, which is a significant improvement compared to the commonly used reacTIVision software.