pith. sign in

arxiv: 2607.01435 · v1 · pith:YT77FAAHnew · submitted 2026-07-01 · 💻 cs.CV · cs.CR· cs.HC· cs.LG

Sign in the Air to Unlock: An Interface for authentication in Virtual and Augmented Reality Powered by Point-Voxel Cross-Attention Network

Neda Abdolrahimi , Thiru Siddharth , Frank Sicongchen , Vir V Phoha This is my paper

Pith reviewed 2026-07-03 20:54 UTC · model grok-4.3

classification 💻 cs.CV cs.CRcs.HCcs.LG
keywords in-air signatureVR authenticationAR authenticationpoint-voxel network3D trajectorybehavioral biometricsimmersive environmentsDeepAirSig
0
0 comments X

The pith

Users authenticate in VR and AR by signing naturally in 3D space, with a point-voxel cross-attention network achieving 2.5 percent equal error rate on one dataset.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces an interface called Sign in the Air to Unlock that lets people authenticate by performing their personal signature as a 3D gesture in virtual or augmented reality. It develops PV-Net, a point-voxel cross-attention network, to process the resulting motion trajectories by combining local dynamics with global spatial structure. Evaluation on the public DeepAirSig collection of 1800 signatures yields a 2.5 percent equal error rate, while a new ImmAirSig set collected in Meta Quest 2 VR shows 76 percent classification accuracy. The approach targets the problem that passwords and external hardware break immersion, while other 3D methods often need specialized sensors. A reader would care because the work tests whether familiar personal gestures can serve as a practical biometric layer inside embodied computing environments.

Core claim

The central claim is that 3D in-air signatures, when modeled by a Point-Voxel Cross-Attention Network that jointly captures local motion dynamics and global spatial structure, supply usable authentication performance in virtual and augmented reality, as shown by the reported error rates on DeepAirSig and the new ImmAirSig dataset collected with consumer VR hardware.

What carries the argument

Point-Voxel Cross-Attention Network (PV-Net) that jointly models local motion dynamics and global spatial structure from 3D trajectories.

If this is right

  • The interface supports natural embodied gestures without requiring external hardware or breaking immersion.
  • It avoids the constraints of eye-tracking or EEG methods that need specialized sensors.
  • Performance on two separate datasets indicates the method can operate in both public and immersive VR settings.
  • It positions 3D behavioral biometrics as a route to merge security with everyday interaction in VR and AR.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the signatures remain stable over longer time gaps than tested, the method could support periodic re-authentication without user friction.
  • The approach might combine with other motion sensors already present in headsets to create multi-factor checks inside the same session.
  • Testing whether the network generalizes to left-handed versus right-handed signing styles would clarify deployment scope.
  • Extending the model to detect fatigue or stress in the signature motion could add secondary safety signals.

Load-bearing premise

That 3D in-air signatures collected from users are sufficiently unique, stable across sessions, and difficult to imitate.

What would settle it

A controlled imitation attack in which an adversary observes and reproduces a target user's 3D signature trajectory and succeeds in authentication above the reported error thresholds across multiple sessions.

Figures

Figures reproduced from arXiv: 2607.01435 by Frank Sicongchen, Neda Abdolrahimi, Thiru Siddharth, Vir V Phoha.

Figure 1
Figure 1. Figure 1: Overview of the PV-Net architecture. The input signing trajectory in 3D space is processed by a dual-branch cross-attention [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Performance curves for the point cloud–only variant of PV-Net on DeepAirSig. [PITH_FULL_IMAGE:figures/full_fig_p010_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Performance curves for the voxel-only variant of PV-Net on DeepAirSig. [PITH_FULL_IMAGE:figures/full_fig_p010_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Performance curves for the full PV-Net model on DeepAirSig (EER = 2.5%). [PITH_FULL_IMAGE:figures/full_fig_p010_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Data collection and signing workflow in the immersive VR environment. [PITH_FULL_IMAGE:figures/full_fig_p011_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Samples of genuine and forged in-air signatures. [PITH_FULL_IMAGE:figures/full_fig_p011_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance curves for the point cloud–only variant of PV-Net on ImmAirSig. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Performance curves for the voxel–only variant of PV-Net on ImmAirSig. [PITH_FULL_IMAGE:figures/full_fig_p013_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Performance curves for the full PV-Net model on ImmAirSig. The model achieves an EER of 24.21%, demonstrating the benefit [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
read the original abstract

Significant advancement of immersive technologies such as Virtual and Augmented Reality (VR/AR) and their integration into diverse aspects of modern life need authentication interfaces that are secure, intuitive, and compatible with embodied interaction. Traditional methods such as passwords, PINs, and device-based logins, break immersion and rely on external hardware. Recent 3D-specific behavioral approaches, such as hand-gesture, eye-tracking, and electroencephalography (EEG)-based methods, offer promising alternatives but often require specialized sensors or constrain natural movement, limiting usability in dynamic environments. We present Sign in the Air to Unlock, an in-air signature interface that enables users to authenticate by signing naturally in 3D space which is a familiar, personal, and reproducible gesture. To realize this interface, we design a point-voxel Cross-Attention Network (PV-Net) that jointly models local motion dynamics and global spatial structure from 3D trajectories. The model is evaluated on two datasets: the public DeepAirSig dataset (1,800 signatures from 40 users) and ImmAirsig, a new dataset collected using Meta Quest 2 in immersive VR (880 samples from 22 users). PV-Net achieves an Equal Error Rate of 2.5% on DeepAirSig and 76% classification accuracy on ImmAirSig. These findings highlight the potential of 3D behavioral interfaces for seamless, user-centric authentication that merges security with natural interaction in immersive environments.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes 'Sign in the Air to Unlock', a 3D in-air signature authentication interface for VR/AR. It introduces PV-Net, a point-voxel cross-attention network to model local motion dynamics and global spatial structure from 3D trajectories, and reports an Equal Error Rate of 2.5% on the public DeepAirSig dataset (1,800 samples from 40 users) together with 76% classification accuracy on the newly collected ImmAirSig dataset (880 samples from 22 users collected with Meta Quest 2).

Significance. If the reported performance is supported by session-respecting splits, temporal stability tests, and imitation-resistance experiments, the work could demonstrate a practical route to hardware-free, embodied authentication that preserves immersion in VR/AR environments.

major comments (2)
  1. [Abstract] Abstract: the headline claims of 2.5% EER on DeepAirSig and 76% accuracy on ImmAirSig are presented without any description of the train/test partitioning strategy, whether splits respect user or session boundaries, or the presence of cross-validation and baseline comparisons; these omissions prevent verification that the numbers establish authentication performance rather than within-user classification.
  2. [Evaluation] Evaluation protocol: no results are supplied on signature stability across temporally separated sessions or on resistance to imitation attacks (e.g., replay from video), yet the abstract frames the EER as evidence for usable authentication security; without these tests the central security claim rests on an unverified assumption.
minor comments (2)
  1. [Abstract] The abstract introduces PV-Net without a one-sentence architectural summary or comparison to prior point-voxel attention models.
  2. [Dataset] Dataset description for ImmAirSig lacks explicit information on the number of sessions per user and the time elapsed between enrollment and verification samples.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We address each major point below and will revise the manuscript to improve clarity on the evaluation protocol while acknowledging limitations in the current security assessment.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the headline claims of 2.5% EER on DeepAirSig and 76% accuracy on ImmAirSig are presented without any description of the train/test partitioning strategy, whether splits respect user or session boundaries, or the presence of cross-validation and baseline comparisons; these omissions prevent verification that the numbers establish authentication performance rather than within-user classification.

    Authors: We agree that the abstract, as a concise summary, should reference the key aspects of the evaluation protocol to support the reported metrics. The full manuscript (Experiments section) specifies user-independent splits that respect session boundaries along with cross-validation and baseline comparisons. We will revise the abstract to include a brief statement on the partitioning strategy and evaluation setup. revision: yes

  2. Referee: [Evaluation] Evaluation protocol: no results are supplied on signature stability across temporally separated sessions or on resistance to imitation attacks (e.g., replay from video), yet the abstract frames the EER as evidence for usable authentication security; without these tests the central security claim rests on an unverified assumption.

    Authors: The referee is correct that the manuscript does not include temporal stability tests across sessions or imitation-resistance experiments, which are relevant for fully substantiating authentication security claims. The current evaluation focuses on performance within the collected datasets. In revision we will add a limitations subsection that explicitly discusses these gaps and their implications for real-world use, while outlining directions for future experiments on stability and imitation attacks. revision: partial

Circularity Check

0 steps flagged

No circularity in derivation chain; results are direct empirical evaluations.

full rationale

The paper reports PV-Net performance as standard ML evaluation outcomes (EER 2.5% on DeepAirSig, 76% accuracy on ImmAirSig) on two named datasets without any claimed derivation, equations, or parameter-fitting steps that reduce to the inputs by construction. No self-citations, uniqueness theorems, or ansatzes are invoked in the provided text to support the architecture or metrics. The security assumptions about signature uniqueness/stability are external premises, not part of any self-referential loop in a derivation chain.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 1 invented entities

The performance claim rests on the domain assumption that 3D hand trajectories constitute usable biometric identifiers and on the empirical results of a newly introduced neural network whose internal parameters are fitted to the signature data.

free parameters (1)
  • PV-Net model parameters
    Weights and hyperparameters of the point-voxel cross-attention network are learned from the signature datasets to produce the reported EER and accuracy.
axioms (1)
  • domain assumption In-air 3D signatures are unique and reproducible enough to serve as authentication tokens
    The entire interface and the meaning of the reported error rates depend on this premise.
invented entities (1)
  • PV-Net no independent evidence
    purpose: Joint modeling of local motion dynamics and global spatial structure from 3D trajectories for signature classification
    New network architecture introduced to realize the authentication interface.

pith-pipeline@v0.9.1-grok · 5822 in / 1339 out tokens · 30795 ms · 2026-07-03T20:54:19.743121+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

41 extracted references · 41 canonical work pages

  1. [1]

    Victoria Olubola Adeyele. 2025. Integrating augmented reality in preschool education: a systematic review. International Journal of Technology Enhanced Learning 17, 3 (2025), 265–284

    work page 2025

  2. [2]

    Anupama Ambika, Hyunju Shin, and Varsha Jain. 2025. Immersive technologies and consumer behavior: A systematic review of two decades of research. Australian Journal of Management 50, 1 (2025), 55–79

    work page 2025

  3. [3]

    Hadi Ardiny and Esmaeel Khanmirza. 2018. The role of AR and VR technologies in education developments: opportunities and challenges. In 2018 6th rsi international conference on robotics and mechatronics (icrom) . IEEE, 482–487

    work page 2018

  4. [4]

    Gonzalo Bailador, Carmen Sanchez-Avila, Javier Guerra-Casanova, and Alberto de Santos Sierra. 2011. Analysis of pattern recognition techniques for in-air signature biometrics. Pattern Recognition 44, 10-11 (2011), 2468–2478

    work page 2011

  5. [5]

    Fadi Boutros, Naser Damer, Kiran Raja, Raghavendra Ramachandra, Florian Kirchbuchner, and Arjan Kuijper. 2020. Iris and periocular biometrics for head mounted displays: Segmentation, recognition, and synthetic data generation. Image and Vision Computing 104 (2020), 104007

    work page 2020

  6. [6]

    Julio Cabero-Almenara, Julio Barroso-Osuna, Carmen Llorente-Cejudo, and María del Mar Fernández Martínez. 2019. Educational uses of augmented reality (AR): Experiences in educational science. Sustainability 11, 18 (2019), 4990

    work page 2019

  7. [7]

    Shu-Hsiang Chen and Fawad Ahmed. 2025. The transformative power of AI and AR in Customer-Centric services in the tourism and hospitality sector: Looking back to predict the future outlook. In Managing customer-centric strategies in the digital landscape . IGI Global, 105–138

    work page 2025

  8. [8]

    Federico De Lorenzis, Alessandro Visconti, Martina Marani, Erik Prifti, Camilla Andiloro, Alberto Cannavò, and Fabrizio Lamberti. 2023. 3DK-reate: Create your own 3D Key for distributed authentication in the Metaverse. In 2023 IEEE Gaming, Entertainment, and Media Conference (GEM) . IEEE, 1–6

    work page 2023

  9. [9]

    Reyhan Duezguen, Peter Mayer, Sanchari Das, and Melanie Volkamer. 2020. Towards secure and usable authentication for augmented and virtual reality head-mounted displays. arXiv preprint arXiv:2007.11663 (2020)

    work page arXiv 2020

  10. [10]

    Oleksii Dzhus and Mykhailo Lobur. 2025. THE INTEGRATION OF ELECTROENCEPHALOGRAPHY AND VIRTUAL REALITY FOR REHABILITA- TION: A PERSPECTIVE REVIEW OF SUCCESSES AND PITFALLS. (2025)

    work page 2025

  11. [11]

    Alain Claude Bah Esseme, Matthew Abiola Oladipupo, Onyedibe Nkiruka Ogechukwu, Nneoma Andrew-Vitalis, Edidiong Elijah Akpan, Victoria En- emona Oseni, and Ugochukwu Okwudili Matthew. 2025. Healthcare applications of augmented reality (AR) and virtual reality (VR): Immersive simulation in medical-clinical education. In Creating Immersive Learning Experienc...

    work page 2025

  12. [12]

    Lei Fan, Junjie Wang, Qi Li, Zhenhao Song, Jinhui Dong, Fangjun Bao, and Xiaofei Wang. 2023. Eye movement characteristics and visual fatigue assessment of virtual reality games with different interaction modes. Frontiers in neuroscience 17 (2023), 1173127

    work page 2023

  13. [13]

    Yuxun Fang, Wenxiong Kang, Qiuxia Wu, and Lei Tang. 2017. A novel video-based system for in-air signature verification. Computers & Electrical Engineering 57 (2017), 1–14

    work page 2017

  14. [14]

    Elyoenai Guerra-Segura, Aysse Ortega-Pérez, and Carlos M Travieso. 2021. In-air signature verification system using leap motion. Expert Systems with Applications 165 (2021), 113797

    work page 2021

  15. [15]

    Hanyang Guo, Hong-Ning Dai, Xiapu Luo, Gengyang Xu, Fengliang He, and Zibin Zheng. 2025. An empirical study on meta virtual reality applications: Security and privacy perspectives. IEEE Transactions on Software Engineering (2025)

    work page 2025

  16. [16]

    Yuheng Guo and Hiroyuki Sato. 2023. Smartwatch In-Air Signature Time Sequence Three-Dimensional Static Restoration Classification Based on Multiple Convolutional Neural Networks. Applied Sciences 13, 6 (2023), 3958

    work page 2023

  17. [17]

    Tommy Hinks, Hamish Carr, Linh Truong-Hong, and Debra F Laefer. 2013. Point cloud data conversion into solid models via point-based voxelization. Journal of Surveying Engineering 139, 2 (2013), 72–83. Manuscript submitted to ACM Sign in the Air to Unlock: An Interface for authentication in Virtual and Augmented Reality Powered by Point–Voxel Cross-Attenti...

    work page 2013

  18. [18]

    Junsik Jung, Han-Cheol Moon, Jooyoung Kim, Donghyun Kim, and Kar-Ann Toh. 2021. Wi-Fi based user identification using in-air handwritten signature. IEEE Access 9 (2021), 53548–53565

    work page 2021

  19. [19]

    Ken Kencevski and Yu Aimee Zhang. 2019. VR and AR for Future Education. In Handbook of mobile teaching and learning . Springer, 1373–1388

    work page 2019

  20. [20]

    Wee How Khoh, Ying Han Pang, Andrew Beng Jin Teoh, and Shih Yin Ooi. 2021. In-air hand gesture signature using transfer learning and its forgery attack. Applied Soft Computing 113 (2021), 108033

    work page 2021

  21. [21]

    Hojoong Kim, Young-Tae Kwon, Hyo-Ryoung Lim, Jong-Hoon Kim, Yun-Soung Kim, and Woon-Hong Yeo. 2021. Recent advances in wearable sensors and integrated functional devices for virtual and augmented reality applications. Advanced Functional Materials 31, 39 (2021), 2005692

    work page 2021

  22. [22]

    Rahul Kumar, Shubhadeep Mukherjee, and Indranil Bose. 2025. Metaverse advertising and promotional effectiveness: The route from immersion to joy. Decision Support Systems 189 (2025), 114386

    work page 2025

  23. [23]

    Varun Kumar and Bijay Prasad Kushwaha. 2025. A systematic literature review of virtual reality in tourism marketing using a mixed method. Cogent Business & Management 12, 1 (2025), 2528161

    work page 2025

  24. [24]

    Gen Li and Hiroyuki Sato. 2022. Sensing in-air signature motions using smartwatch: A high-precision approach of behavioral authentication. IEEE Access 10 (2022), 57865–57879

    work page 2022

  25. [25]

    Feng Lin, Kun Woo Cho, Chen Song, Wenyao Xu, and Zhanpeng Jin. 2018. Brain password: A secure and truly cancelable brain biometrics for smart headwear. In Proceedings of the 16th Annual International Conference on Mobile Systems, Applications, and Services . 296–309

    work page 2018

  26. [26]

    Jameel Malik, Ahmed Elhayek, Sheraz Ahmed, Faisal Shafait, Muhammad Imran Malik, and Didier Stricker. 2018. 3dairsig: A framework for enabling in-air signatures using a multi-modal depth sensor. Sensors 18, 11 (2018), 3872

    work page 2018

  27. [27]

    Jameel Malik, Ahmed Elhayek, Suparna Guha, Sheraz Ahmed, Amna Gillani, and Didier Stricker. 2020. DeepAirSig: End-to-end deep learning based in-air signature verification. IEEE Access 8 (2020), 195832–195843

    work page 2020

  28. [28]

    Robert Miller, Natasha Kholgade Banerjee, and Sean Banerjee. 2020. Within-system and cross-system behavior-based biometric authentication in virtual reality. In 2020 IEEE conference on virtual reality and 3D user interfaces abstracts and workshops (VRW) . IEEE, 311–316

    work page 2020

  29. [29]

    Naheem Noah and Sanchari Das. 2025. From PINs to Gestures: Analyzing Knowledge-Based Authentication Schemes for Augmented and Virtual Reality. IEEE Transactions on Visualization and Computer Graphics (2025)

    work page 2025

  30. [30]

    Omid Panahi. 2025. Innovations in Surgical Healthcare: The Future of Medicine. Journal of Surgical Case Reports and Investigations 1, 1 (2025), 1–8

    work page 2025

  31. [31]

    Jafar Pourbemany, Ye Zhu, and Riccardo Bettati. 2023. A survey of wearable devices pairing based on biometric signals. IEEE Access 11 (2023), 26070–26085

    work page 2023

  32. [32]

    Mamtha Prajapati and Sudesh Kumar. 2025. Virtual reality revolution in healthcare: a systematic review. Health and Technology 15, 2 (2025), 231–242

    work page 2025

  33. [33]

    Bhupinder Singh and Saurabh Chandra. 2025. Augmented Reality and Virtual Reality in Health Advancements and Medical Education: Runway for Future Ready Healthcare Services. In Digital Twins for Sustainable Healthcare in the Metaverse . IGI Global Scientific Publishing, 197–218

    work page 2025

  34. [34]

    Sophie Stephenson, Bijeeta Pal, Stephen Fan, Earlence Fernandes, Yuhang Zhao, and Rahul Chatterjee. 2022. Sok: Authentication in augmented and virtual reality. In 2022 IEEE symposium on security and privacy (SP) . IEEE, 267–284

    work page 2022

  35. [35]

    Yi Tan, Wenyu Xu, Shenghan Li, and Keyu Chen. 2022. Augmented and virtual reality (AR/VR) for education and training in the AEC industry: A systematic review of research and applications. Buildings 12, 10 (2022), 1529

    work page 2022

  36. [36]

    Yangyi Eric Tang and Qi Zhou. 2025. Inspired by intertemporal connections: Using AR technology to enhance visitor satisfaction in historical museums. Tourism management 108 (2025), 105096

    work page 2025

  37. [37]

    Jiawei Wang, BoYu Gao, Huawei Tu, Hai-Ning Liang, Zitao Liu, Weiqi Luo, and Jian Weng. 2024. Secure and memorable authentication using dynamic combinations of 3d objects in virtual reality. International Journal of Human–Computer Interaction 40, 17 (2024), 4608–4626

    work page 2024

  38. [38]

    Waqas Wazir, Hasan Ali Khattak, Ahmad Almogren, Mudassar Ali Khan, and Ikram Ud Din. 2020. Doodle-based authentication technique using augmented reality. IEEE Access 8 (2020), 4022–4034

    work page 2020

  39. [39]

    Dhruv Kumar Yadav, Beatrice Ionascu, Sai Vamsi Krishna Ongole, Aditi Roy, and Nasir Memon. 2015. Design and analysis of shoulder surfing resistant pin based authentication mechanisms on google glass. In Financial Cryptography and Data Security: FC 2015 International Workshops, BITCOIN, W AHC, and Wearable, San Juan, Puerto Rico, January 30, 2015, Revised ...

    work page 2015

  40. [40]

    Zhen Yu, Hai-Ning Liang, Charles Fleming, and Ka Lok Man. 2016. An exploration of usable authentication mechanisms for virtual reality systems. In 2016 IEEE Asia Pacific Conference on Circuits and Systems (APCCAS) . IEEE, 458–460

    work page 2016

  41. [41]

    Yongtuo Zhang, Wen Hu, Weitao Xu, Chun Tung Chou, and Jiankun Hu. 2018. Continuous authentication using eye movement response of implicit visual stimuli. proceedings of the acm on interactive, mobile, wearable and ubiquitous technologies 1, 4 (2018), 1–22. Received 20 February 2007; revised 12 March 2009; accepted 5 June 2009 Manuscript submitted to ACM

    work page 2018