Vision AI or Telematics AI — which does the job?
The default assumption in industrial AI is that computer vision can answer any field question — given enough cameras, enough compute, and enough training data. That assumption breaks at the first roller pass count, the first fuel reconciliation, and the first proximity event in low-visibility conditions.
Visual detection answers the question 'what is happening in this frame?' very well. It does not answer 'what has happened across this asset for the last 8 hours' — at least not without telemetry as a foundation. Vision AI without Physics AI is observational. Physics AI without Vision AI is partial. The two together cover what neither can alone.
- Vision AI strength: rich event detection — PPE, intrusion, fire, defect classification
- Vision AI weakness: cumulative state — pass counts, fuel reconciliation, cycle history
- Physics AI strength: continuous telemetry — pass counts, fuel flow, GPS, sensor state
- Physics AI weakness: visual judgement — spread quality, defect classification, PPE compliance
Two layers, one Cortex knowledge graph.
Cognecto's Boson layer ingests telematics from 20+ OEM devices through a unified codec library — normalising every protocol into one schema. Photon's vision intelligence ingests RTSP/RTMP camera feeds and applies trained models per use case (PPE, intrusion, fire, spread quality, defect classification).
Both data layers feed the same Cortex knowledge graph. A roller pass detected by Boson and the cement spread verified by Photon both clear the same BOQ line item. A driver fatigue event from Photon's DMS correlates with the GPS overspeed event from Boson's telematics — same incident, two streams, one investigation.
Where each AI carries the work.
Compaction sequence verification: Boson (pass count, GPS, sequence). Cement spread uniformity: Photon (vision). PPE compliance: Photon. Fuel reconciliation: Boson. Equipment idle: both. Driver fatigue: Photon. Geofence breach: Boson. Fire detection: Photon. Predictive maintenance: Boson. No-go zone intrusion: Photon. The list goes on. The principle is constant: use the AI that matches the physical reality.