Video, image and biometric data

Drometra Vision is currently a closed beta program. Participants are companies that expressly request it, with a compatible Android phone installed in the vehicle and the camera pointed at the road. The dedicated Drometra Vision camera is in development and is not yet commercially available.

The signals produced by the beta —detection of people and vehicles ahead of the vehicle— are experimental. They are not safety alerts, do not act on the vehicle, and must not be used as the sole basis for evaluating anyone's driving.

• Manual mode: the user starts and stops recording; each clip lasts at most 119 seconds.
• Workday mode: records in 119-second segments for a maximum of 8 hours, and stops when finished.
• Events mode: keeps a circular buffer of a few seconds and retains only a context window around each detection.

The detection model is EfficientDet-Lite0, an open model (Apache 2.0 license) that runs on the phone's processor through Google's MediaPipe library. It was not trained on images from customers or beta participants.

Biometric data is data that makes it possible to identify a person from their physical features: a facial vector, a fingerprint, their gait. Detecting that there is a person on the roadway is not biometric data, as long as it is not used to find out who they are. Drometra Vision is limited to the latter: it knows there is «a person» or «a vehicle», not who.

If in the future we offer features that process biometric data or that are directed at the cab —for example, fatigue detection— we will do so only after: a data protection impact assessment; a labor review in accordance with the doctrine of the Labor Directorate (Dirección del Trabajo); a reinforced lawful basis, which for sensitive data requires express consent or legal authorization; and the prior update of this document.

A camera on the road captures pedestrians, other drivers, patents and facades. These people have no relationship with our client and did not choose to be recorded, leaving them in the most unprotected position of the entire system.

• We do not apply facial recognition to the footage, in any configuration.
• We do not read license plates or build license plate databases.
• Detections use temporary identifiers that are discarded when the object leaves the frame; no one is tracked across clips or across days.
• We recommend that the participant disclose the use of the camera with a visible sticker on the vehicle.

• On the phone: when each clip is closed, it is encrypted with AES-256-GCM using a key stored in Android's secure store (Android Keystore). While recording, the file is protected by the system's own encryption. The space used has a configurable cap (8 GiB by default).
• Cloud upload: only happens if the participant expressly enables automatic upload. Clips are transferred to lab object storage on Cloudflare R2.
• Lab use: the cloud copy is used to validate detection quality during the beta. The production version, with per-device identity and per-company separation, is in development.

1. 1.On the phone, the material is accessible only from the app, with the session of the user authorized by the participating company.
2. 2.In the lab cloud, only the Drometra technical team assigned to the beta, to validate detection and resolve issues at the participant's request.
3. 3.We do not share the footage with third parties, except when required by a competent authority in accordance with our authority request policy.

Once the period ends, the material is deleted automatically. By default the app does not delete a local clip that could not be uploaded, so as not to lose evidence; the participant can enable local deletion once the cloud copy has been verified. Uninstalling the app deletes the encryption keys and, with them, the ability to read any clips remaining on the phone.