TEMPEST
A driver identification framework based on temporal convolutional networks and ArcFace additive angular margin loss. It encodes behavioral biometrics into 96-dimensional embedding vectors to enable scalable identity recognition, supporting dynamic enrollment of new drivers without retraining the model.
Timeline (last 90 days)
The core of TEMPEST is a temporal convolutional network (TCN) embedding model trained with an additive angular margin loss (ArcFace loss).
TEMPEST maps a 60-second multimodal driving window to a 96-dimensional embedding vector.
TEMPEST supports dynamic enrollment: when new drivers are added to the fleet, there is no need to retrain the model or refit the classifier — enrollment is completed simply by extracting embeddings.
On a dataset containing 45 drivers, TEMPEST achieves a Rank-1 accuracy of 91.71% under strict temporal evaluation.
TEMPEST outperforms the best classical model by 17.9 percentage points and the strongest triplet-loss baseline by 58.4 percentage points.
When the driver pool expands from 10 to 45 drivers, TEMPEST drops by only 4.3 percentage points, while the supervised triplet-loss baseline drops by 22 percentage points and the unsupervised triplet-loss baseline drops by 32.5 percentage points.
On the public KIA Soul dataset, TEMPEST outperforms the best classical model by 7.3 percentage points within-session and by 14.3 percentage points cross-session.
The TEMPEST model has only 720,000 parameters, a storage footprint of 2.80 MB, and is able to converge within 50 training epochs.
The paper positions TEMPEST as a rigorous and reproducible baseline in the field of scalable behavioral driver biometrics.
The TEMPEST paper does not deeply discuss robustness in adversarial scenarios (such as deliberately mimicking another person's driving style), and the scale of 45 drivers remains limited relative to actual large fleets.
All Facts (10)
The core of TEMPEST is a temporal convolutional network (TCN) embedding model trained with an additive angular margin loss (ArcFace loss).
50%UnverifiedTEMPEST maps a 60-second multimodal driving window to a 96-dimensional embedding vector.
50%UnverifiedTEMPEST supports dynamic enrollment: when new drivers are added to the fleet, there is no need to retrain the model or refit the classifier — enrollment is completed simply by extracting embeddings.
50%UnverifiedOn a dataset containing 45 drivers, TEMPEST achieves a Rank-1 accuracy of 91.71% under strict temporal evaluation.
50%UnverifiedTEMPEST outperforms the best classical model by 17.9 percentage points and the strongest triplet-loss baseline by 58.4 percentage points.
50%UnverifiedWhen the driver pool expands from 10 to 45 drivers, TEMPEST drops by only 4.3 percentage points, while the supervised triplet-loss baseline drops by 22 percentage points and the unsupervised triplet-loss baseline drops by 32.5 percentage points.
50%UnverifiedOn the public KIA Soul dataset, TEMPEST outperforms the best classical model by 7.3 percentage points within-session and by 14.3 percentage points cross-session.
50%UnverifiedThe TEMPEST model has only 720,000 parameters, a storage footprint of 2.80 MB, and is able to converge within 50 training epochs.
50%UnverifiedThe paper positions TEMPEST as a rigorous and reproducible baseline in the field of scalable behavioral driver biometrics.
50%UnverifiedThe TEMPEST paper does not deeply discuss robustness in adversarial scenarios (such as deliberately mimicking another person's driving style), and the scale of 45 drivers remains limited relative to actual large fleets.
50%