TEMPEST: Temporal Embeddings for Scalable Driver Identification via Angular Margin Learning

Computer Science > Machine Learning arXiv:2610.06855 (cs) [Submitted on 20 May 2026] Title: TEMPEST: Temporal Embeddings for Scalable Driver Identification via Angular Margin Learning Authors: Kyle Musgrove, Dylan B. Lewis, Sarah Powers, Emma J. Reid, Hector Santos-Villalobos.

计算机科学 > 机器学习 arXiv:2610.06855 (cs) [2026年5月20日提交] 标题：TEMPEST：通过角度边缘学习实现可扩展驾驶员识别的时间嵌入。作者：Kyle Musgrove, Dylan B. Lewis, Sarah Powers, Emma J. Reid, Hector Santos-Villalobos。

Abstract: Scalable driver identification requires embedding models that maintain discriminative performance as fleet size grows, yet existing triplet-loss formulations degrade rapidly with driver pool size and overfit to session-specific patterns under rigorous temporal evaluation. We introduce TEMPEST, a Temporal Convolutional Network embedding model trained with an additive angular margin (ArcFace) loss that enforces global class-level separation in a normalized angular space.

摘要：可扩展的驾驶员识别需要嵌入模型在车队规模扩大时保持判别性能，然而现有的三元组损失公式在驾驶员群体规模扩大时性能迅速下降，且在严格的时间评估下容易过拟合特定会话模式。我们引入了 TEMPEST，这是一种时间卷积网络嵌入模型，通过加性角度边缘（ArcFace）损失进行训练，在归一化的角度空间中强制实现全局类级分离。

TEMPEST maps 60-second multimodal driving windows to compact 96-dimensional embeddings, supporting truly dynamic enrollment without any retraining or classifier refitting. Under rigorous temporal evaluation on a 45-driver dataset, TEMPEST achieves 91.71% Rank-1 accuracy, outperforming the best classical model by 17.9 pp and the strongest triplet-loss baseline by 58.4 pp.

TEMPEST 将 60 秒的多模态驾驶窗口映射为紧凑的 96 维嵌入，支持真正的动态注册，无需任何重新训练或分类器调整。在针对 45 名驾驶员数据集的严格时间评估中，TEMPEST 达到了 91.71% 的 Rank-1 准确率，比最佳经典模型高出 17.9 个百分点，比最强的三元组损失基线高出 58.4 个百分点。

TEMPEST degrades by only 4.3 pp when growing the subject pool from 10 to 45 drivers, compared to 22 pp and 32.5 pp for supervised and unsupervised triplet-loss baselines, and its cross-session advantage is corroborated on the public KIA Soul dataset, where it outperforms the best classical model by 7.3 pp within-session and 14.3 pp cross-session. With 720K parameters, a 2.80 MB footprint, and 50-epoch convergence, TEMPEST establishes a rigorous, reproducible baseline for scalable behavioral driver biometric identification.

当受试者群体从 10 名增加到 45 名时，TEMPEST 的性能仅下降了 4.3 个百分点，而监督和无监督三元组损失基线分别下降了 22 个百分点和 32.5 个百分点。其跨会话优势在公开的 KIA Soul 数据集上得到了证实，在该数据集上，它在会话内比最佳经典模型高出 7.3 个百分点，在跨会话中高出 14.3 个百分点。凭借 72 万个参数、2.80 MB 的占用空间以及 50 个周期的收敛速度，TEMPEST 为可扩展的行为驾驶员生物识别建立了一个严格且可复现的基准。