Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing 3D-Air Signature by Pen Tip Tail Trajectory Awareness: Dataset and Featuring by Novel Spatio-temporal CNN

Saurabh Atreya, Maheswar Bora, Aritra Mukherjee, Abhijit Das

TL;DR

This work tackles air-signature biometrics by exploiting the joint 3D trajectories of pen tip and tail captured with a stereo camera. It introduces SliTCNN, a two-stream 2D spatio-temporal CNN, and the T3AAS-v1 dataset to capture and benchmark genuine versus forged air signatures. Across multiple datasets, the two-stream SliTCNN demonstrates superior recognition and verification performance with a compact model, outpacing common temporal models like LSTM/GRU and 1D-CNN, and approaching VGG16 accuracy with far fewer parameters. The results underscore the value of incorporating depth- and orientation-rich tip-tail dynamics for robust, edge-friendly air-signature biometrics with clear practical implications for AR/VR and real-time authentication.

Abstract

This work proposes a novel process of using pen tip and tail 3D trajectory for air signature. To acquire the trajectories we developed a new pen tool and a stereo camera was used. We proposed SliT-CNN, a novel 2D spatial-temporal convolutional neural network (CNN) for better featuring of the air signature. In addition, we also collected an air signature dataset from $45$ signers. Skilled forgery signatures per user are also collected. A detailed benchmarking of the proposed dataset using existing techniques and proposed CNN on existing and proposed dataset exhibit the effectiveness of our methodology.

Enhancing 3D-Air Signature by Pen Tip Tail Trajectory Awareness: Dataset and Featuring by Novel Spatio-temporal CNN

TL;DR

This work tackles air-signature biometrics by exploiting the joint 3D trajectories of pen tip and tail captured with a stereo camera. It introduces SliTCNN, a two-stream 2D spatio-temporal CNN, and the T3AAS-v1 dataset to capture and benchmark genuine versus forged air signatures. Across multiple datasets, the two-stream SliTCNN demonstrates superior recognition and verification performance with a compact model, outpacing common temporal models like LSTM/GRU and 1D-CNN, and approaching VGG16 accuracy with far fewer parameters. The results underscore the value of incorporating depth- and orientation-rich tip-tail dynamics for robust, edge-friendly air-signature biometrics with clear practical implications for AR/VR and real-time authentication.

Abstract

This work proposes a novel process of using pen tip and tail 3D trajectory for air signature. To acquire the trajectories we developed a new pen tool and a stereo camera was used. We proposed SliT-CNN, a novel 2D spatial-temporal convolutional neural network (CNN) for better featuring of the air signature. In addition, we also collected an air signature dataset from signers. Skilled forgery signatures per user are also collected. A detailed benchmarking of the proposed dataset using existing techniques and proposed CNN on existing and proposed dataset exhibit the effectiveness of our methodology.
Paper Structure (17 sections, 7 figures, 5 tables)

This paper contains 17 sections, 7 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Proposed methodology for Air Signature
  3. Proposed air signature capturing set-up
  4. The stereo camera
  5. The pen tool
  6. The capturing setup and protocols
  7. Pre-processing of air signature
  8. SliTCNN for Air Signature
  9. Experimental results
  10. Detailed Description of the Proposed Datasets
  11. Other Datasets used for Comparison
  12. Implementation details
  13. Benchmarking techniques
  14. Results and discussion
  15. Benchmarking results
  16. ...and 2 more sections

Figures (7)

  • Figure 1: Overall pipeline of proposed methodology. SVO is the file format of calibrated stereo video by Zed
  • Figure 2: Schematic diagram of the signature capturing process. The frames above show the final pen position (no transparency) as seen by the stereo cameras.
  • Figure 3: The proposed capturing setup and a screenshot of the software (a)Entire setup for air signature acquisition (b)Screenshot of capturing software
  • Figure 4: The different forms of signature used in experiments: (a)interpolated 2D tip trace (b)interpolated 3D tip trajectory in red with the sampled positions in blue stars (c)interpolated 3D tip and tail trajectory with tip in red and tail in green
  • Figure 5: The detailed network architecture of the proposed 2-stream SliTCNN.
  • ...and 2 more figures