Research map / Architectures and efficiency / RNNs and normalization
Rider Posture-Based Continuous Authentication with Few-Shot Learning for Mobility Scooters (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a posture-based continuous authentication system for mobility scooter riders using spatio-temporal graph convolutions with hierarchical body-part encoders trained by triplet loss, and releases a public keypoint dataset from 42 riders.
Problem
Mobility scooter riders, especially seniors, need seamless continuous authentication, since physical keys and password schemes are inconvenient and existing continuous authentication methods suffer from comfort, cost and long enrollment issues.
Approach
Uses upper-body keypoint sequences from user-facing cameras (no face) processed by spatio-temporal graph convolutions and four body-part residual encoders, producing a 64-d rider embedding trained with triplet metric loss and triplet mining, so that authentication needs only a few minutes of enrollment.
Key results
- MoveNet-based system reaches ROC AUC 0.990 with 10 enrollment samples and 0.990 with 40.
- MediaPipe-based system reaches ROC AUC 0.981 with 1 sample but drops to 0.896 with 5 samples, showing higher variability than MoveNet.
- Dataset comprises 42 participants and about 10 hours 9 minutes (1.1 million frames) of riding footage.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use MoveNet instead of MediaPipe for pose estimation | Mobility scooter authentication system · Mobility scooter authentication across 1, 5, 10, 20, and 40 enrollment samples | ROC AUC: 0.981 vs 0.976; 0.896 vs 0.986; 0.962 vs 0.990; 0.960 vs 0.987; 0.964 vs 0.990 | Mixed |
| Increase enrollment samples from 1 to 10 | Mobility scooter authentication system · Mobility scooter authentication | ROC AUC: MoveNet: 0.976 -> 0.990; MediaPipe: 0.981 -> 0.962 | Mixed |
Surprising findings
- The MediaPipe-based system scores highest with a single enrollment sample (0.981) but is unstable at 5 samples (0.896), trailing MoveNet overall.
Limitations
- Small participant pool (42, mostly 18-25 year olds on campus) with few senior riders (4 over 60).
- Evaluation reported only as ROC AUC on in-house data; no comparison with other continuous authentication methods.
Experimental setup
Datasets: mobility scooter rider video collected from 42 participants on campus
Benchmarks: mobility scooter rider dataset (in-house)
Metrics: ROC AUC
Baselines: MediaPipe MoveNet
Builds on: triplet metric loss spatio-temporal graph convolutional networks
Closest papers by meaning
- TraXion: Rethinking Pre-training Frameworks for Mobility and Beyond (NeurIPS 2026 · similarity 0.79)
- SafeAuto: Knowledge-Enhanced Safe Autonomous Driving with Multimodal Foundation Models (ICML 2025 · similarity 0.77)
- Temporal Logic-Based Multi-Vehicle Backdoor Attacks against Offline RL Agents in End-to-end Autonomous Driving (NeurIPS 2025 · similarity 0.77)
- PID: Prompt-Independent Data Protection Against Latent Diffusion Models (ICML 2024 · similarity 0.76)
- Correlation-Attention Masked Temporal Transformer for User Identity Linkage Using Heterogeneous Mobility Data (AAAI 2025 · similarity 0.76)
- Structural Information Guided Multimodal Pre-training for Vehicle-Centric Perception (AAAI 2024 · similarity 0.76)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.