Unveiling Large-Scale Historical Contents With v-Slam and Markerless Mobile Ar Solutions

Abstract

Abstract. Augmented Reality (AR) is already transforming many fields, from medical applications to industry, entertainment and heritage. In its most common form, AR expands reality with virtual 3D elements, providing users with an enhanced and enriched experience of the surroundings. Until now, most of the research work focused on techniques based on markers or on GNSS/INS positioning. These approaches require either the preparation of the scene or a strong satellite signal to work properly. In this paper, we investigate the use of visual-based methods, i.e., methods that exploit distinctive features of the scene estimated with Visual Simultaneous Localization and Mapping (V-SLAM) algorithms, to determine and track the user position and attitude. The detected features, which encode the visual appearance of the scene, can be saved and later used to track the user in successive AR sessions. Existing AR frameworks like Google ARCore, Apple ARKit and Unity AR Foundation recently introduced visual-based localization in their frameworks, but they target mainly small scenarios. We propose a new Mobile Augmented Reality (MAR) methodology that exploits OPEN-V-SLAM to extend the application range of Unity AR Foundation and better handle large-scale environments. The proposed methodology is successfully tested in both controlled and real-case large heritage scenarios. Results are available also in this video: https://youtu.be/Q7VybmiWIuI.

Benchmarking Built-In Tracking Systems for Indoor AR Applications on Popular Mobile Devices

“…In particular, in addition to the inertial data, visual natural feature points in the environment are recognized and extracted from the captured images, and then used to estimate the position and orientation of a device in the real world [ 51 ]. As a consequence, the device’s ability to recognize and acquire the natural features of the environment is a key aspect that can affect its tracking performance [ 52 ]. Therefore, the following table ( Table 5 ) shows the average number of feature points detected by the selected devices on the three paths; specifically, the number of reference points detected while covering the same path ten times, in the same experimental environment and under the same lighting conditions.…”

Photogrammetry Now and Then – From Hand-Crafted to Deep-Learning Tie Points –

“…1 depict the main square of Trento (Italy). The historical scans are from the "TOTEM" project (Torresani et al, 2021), and have a limited image resolution (972x1364 px, 679x512 px, 713x512 px, 1024x689 px, and 984x1230 px), while the recent images have been collected for this work and have both a resolution of 1500x1000 px (actually, they have been down-sampled from the original acquisition resolution for computational reasons). In addition to the first image pairs, these ones contain also relevant scale and viewpoint variations.…”

“…Moreover, the printing process from the original film and the digitization of the prints or of the original film itself by using a flatbed scanner or by photographing the hardcopies is likely to introduce additional geometric deformations with complex mathematical modelling (Nocerino et al, 2012a). Multi-temporal image co-registrations can be useful for multiple purposes including, but not limited to, Augmented and Virtual Reality (AR/VR) applications (Torresani et al, 2021;Maiwald et al, 2019b), the valorisation of archival photos (Nocerino et al, 2012b), 3D reconstruction of destroyed building facades (Brauer-Burchardt and Voss, 2002) or statues (Gruen et al, 2004), and environmental and climate changes monitoring (Holmlund, 2021). When dealing with multi-temporal datasets, the approaches are different depending on the number of images per epoch.…”

“…Classic hand-crafted local features can sometimes be used when it is not possible to rely on the reconstructed scene of single epochs (Ali and Whitehead, 2014;Maiwald, 2019a), but the extracted correspondences are usually insufficient, and a high number of outliers is present. In this case traditional descriptors generally fail (Farella et al, 2022), and tie points must be collected manually (Holmlund, 2021;Torresani et al, 2021). Recently, Maiwald (2019a) proposed a benchmark for historical images, analyzing popular hand-crafted local features such as ORB+SURF, the Maximally Stable Extremal Regions (MSER, Matas et al, 2004) in conjuction with SURF, and the Radiation-Invariant Feature Transform (RIFT, Li et al, 2018).…”

3D Surveying of Underground Built Heritage: Opportunities and Challenges of Mobile Technologies

“…The 3D digitization of Cultural Heritage (CH) is also considered a common practice for documentation, conservation, preservation, valorization, and visualization purposes [1][2][3]. Moreover, 3D data are an efficient medium for tourist attractions [4,5], for digital archiving and dissemination of artefacts and monuments to future generations [6][7][8], or for allowing VR/AR (Virtual Reality/Augmented Reality) access with mobile devices [9][10][11][12][13][14].…”