Spatial Data Comparison for Markerless Augmented Reality (AR) Anchoring

A rapid growth of Augmented Reality (AR) in recent years can provide a solution to various applications. AR technology can contribute to efficient facility and machinery maintenance and repair by offering holographic overlays and providing information for the current location including manual, specification, revision history and others. The most challenging task for AR systems is the accurate localization of the AR device. Localization refers to the evaluation of the pose information including position and orientation of the viewpoint. The real-time localization problem has been extensively investigated with different approaches in recent years for its key role played in AR applications. Localization is possibly performed with different sensors including indoor and outdoor Global Positioning System (GPS), Wireless Local Area Networks (WLAN), Radio Frequency Identification (RFID). These sensors require a costly equipment infrastructure for readers and tags. A computer vision-based localization approach can be accurate and robustness against sensor-based approaches. Although marker-based approach appears to be the most favorable among the standard vision techniques currently applied in many commercial AR applications, it is time-consuming and infeasible to install markers in a larger space and on each equipment. In addition, these fiducial markers possibly have some esthetical issues in some applications. As an alternative, image-based methods including image-retrieval approaches have been widely used to AR application for building and facility maintenance with making use Building Information Modeling (BIM). Image-retrieval approach is based on finding the most similar image to the queried image among the dataset of images by comparing the features from images. Various ways to extract visual features are adopted to the image comparison such as Scale-Invariant Feature Transform (SIFT) and Speeded Up Robust Features (SURF). The pose information of the viewpoint can be estimated from the closest image because the corresponding information is known from the pre-collected dataset. The dataset images can be preliminarily used by image or rendered from a digital twin at an earlier time. Rendered images from the CAD model can be used as a dataset and the pose information including position and orientation of each image can be easily identified. This approach finds the closest image from the dataset to the image from the AR user’s viewpoint.