Corner-based Square Fiducial Marker Detection for Hand-manipulated AR Objects

Abstract

Square fiducial markers are commonly used in Augmented Reality (AR) applications to affix AR content to a particular location in the real world. Unoccluded, these markers are quickly and easily identified, and AR content is realistically displayed in real time. However, because most square fiducial marker libraries use a thresholding-based method of detection, small edge occlusions often prevent markers from being found, or cause inaccurate estimations of marker pose. Both of these scenarios result in visual disturbances in the AR content. This is particularly problematic for hand-manipulated AR objects, where markers will suffer frequent edge occlusions by fingers. In this paper, we propose an alternative method of detecting single square fiducial markers where only two diagonal corners of the marker must be visible for detection. Our proposed method finds and classifies corners in the image, pairs candidate diagonal corners based on their gradient directions, and then attempts to find a homography between a standard template and corners in the image that may belong to a marker. Identification of potential markers is done using a commonly-used square fiducial marker identification algorithm. This method detects markers under partial edge occlusions at a rate of up to 2.48x that of a popular square fiducial library, detects and localizes square fiducial markers in isolated frames, and is fast enough to be used for real-time applications.