VLM output
An overhead camera watches a wide, nearly empty multi-lane road divided by a double solid center line. A single silver sedan enters from the lower right, drives diagonally across the lanes and over the solid median line, and ends up traveling in the opposite carriageway heading the other way, completing a U-turn across the divider.
- Violation DetectedYes
- Violation TypeIllegal U-Turn
- Solid Line CrossingYes
- Crossed Lanes DiagonallyYes
- Vehicles Involved1
- Vehicle TypeSedan
- Other Traffic PresentNo
- Pedestrians PresentNo
- Traffic DensityVery Light
Customer challenges
One dedicated model per violation type; a scene with several violations is not read as a whole
Legacy AI matches objects; it cannot reason across signal state, markings, trajectory and time
Existing intersection footage is recorded but never analysed for violations
Adding a new violation type means retraining and re-deploying, 12–18 months
No violation pattern or hotspot analysis to guide engineering and campaigns
Customer benefits
5–6x coverage on the same budget; drivers see enforcement everywhere, deterrence spreads from points to the whole network
One camera, many functions, central inference: hardware is roughly 1/N of legacy systems and total cost of ownership drops sharply
New violation types go live in 2–4 weeks instead of a 12–18-month procurement cycle
Equipment can be rotated to high-crash intersections; the total count stays fixed while always pointing at hotspots
The traffic bureau gets city-wide violation pattern and hotspot data for the first time, grounding engineering and campaigns
Performance measured by crash prevention (A1 and A30 reductions) rather than ticket counts, in line with road-safety policy
Key features
Lane Compliance
Detects lane compliance violations.
Parking Violations
Detects vehicles parked in violation.
Speed & Motion Monitoring
Monitors vehicle speed and movement.
Before & after
| Before | After — with Linker AI Nexa |
|---|---|
| Before Each camera analysed for 0–1 violation types | After — with Linker AI Nexa 6–8 violation types per camera, detection scope per unit 6–8x |
| Before Violations reviewed only when a complaint is filed | After — with Linker AI Nexa 100% of in-view violations logged with GPS-synced timestamp and plate |
| Before A new violation type takes 12–18 months to deploy | After — with Linker AI Nexa 2–4 weeks city-wide, deployment lead time -95% |
| Before About US$1,500 per monitored point per month | After — with Linker AI Nexa About US$250 per monitored point per month, unit cost -83% |
| Before Full-frame inference on every camera | After — with Linker AI Nexa Deep inference on 5–10% of frames, compute demand -90% |
| Before Violation data limited to ticket counts | After — with Linker AI Nexa 100% of events geo- and time-stamped for hotspot analysis, refreshed daily |