Vision Lab
An experimental capacity is not announced as a proven delivery. This is the way to a detection before we sell it.
Stages and required evidence
Each stage has a concrete work and an output condition. It is passed to the next one when there is the evidence that the stage asked for, even if the calendar says otherwise.
- V0In progress
Physical feasibility
Candidate device, assembly, energy, capture, codec, buffer and manual event.
To continue: Test in days and conditions representative of what is offered.
- V1Pending
Experimental road
A small vehicle and person detector, tracking and profiling, without critical alarm to the driver.
To continue: Latency, frames per second held, temperature, memory, storage and battery with the exact device and model.
- V2Pending
Event platform
Code roll, structured event, realizable rise, timeline, review, retention and audited access.
To continue: Correlated event and clip; reboot and network lack do not duplicate; demonstrated expiry and recovery.
- V3Pending
Quality of the case of use
Authorized labelling, negative scenarios, day and night, and mounting conditions.
To continue: Precision and completeness by type of event, false positive per hour and capture failures, with sample size and uncertainty. A demonstration video is not enough.
- V4Pending
Commercial pilot
Tackled delivery on accepted devices, with installation and support.
To continue: Agreed representative observation, known cost and a paying customer. Unproven conditions remain excluded.
- V5Pending
Cabin and two nodes
Work and privacy assessment, interior capture and synchronization between nodes.
To continue: Validation by lighting, person and device; access policy; gap measured between nodes.
- V6Pending
Enlargement
Recording of variants, cohort deployment, more devices and accelerator optimization.
To continue: Sustained quality, model reversion and favourable cost against alternatives.
- V7Pending
Own Hardware
Node dedicated with automotive energy, infrared lighting and necessary approvals.
To continue: Demanda suficiente y costo total documentado.
Licences before capabilities
Before selecting a model we check the license of the code, weights and datasets, in addition to the rights of redistribution, commercial use and attribution obligations. That a repository is public or that there is an academic article does not automatically grant these rights.
Detecting an object and detecting an incident are different assessments. Seeing many vehicles does not justify uploading a clip for each one: temporary window, cooling and context signals are applied to control noise and cost.
The evaluation of patents, the precise monocular distance, the universal detection of lanes, night somnolence and the use of infrared lighting are explicitly in place. Audio, facial recognition and biometrics are disabled in the initial range.
Preguntas frecuentes
Do you have a specific use case?
The tests are prioritized with real cases. Tell us what you need to detect and under what conditions your fleet operates.