Measure Road Sentinel against the road.
The verification plan compares Road Sentinel observations with known vehicle runs, surveyed locations, time references, and human counts. Results will be published by field and condition as validation proceeds.
A car we already know
Every new mile
We will drive the mile with a GPS logger bolted to the roof, then line our record up against the logger run by run.
Fails at
The run falls outside the acceptance threshold set for that validation plan.
A tape measure
At install
Someone will walk the spacing with a measuring wheel and write it on paper, and the paper is what the pylons get compared against.
Fails at
The installed position does not match the surveyed site record.
The clock
Hourly, automatic
Each pylon is designed to check its own clock against GPS time and log the drift, whether or not anyone is looking.
Fails at
Clock drift exceeds the threshold set for the data use.
A human with a stopwatch
Monthly
A person stands at the roadside for ten minutes and counts vehicles by hand. We compare that count with what the pylon reported for the same ten minutes.
Fails at
Counts or classes fall outside the acceptance threshold for that test.
A failed check remains in the validation record with its reason and the affected data fields.
Publish a result for each condition.
A prototype is built, and field validation comes next. Each row below names a result the program will publish instead of combining every condition into one performance number.
| Field or condition | Where it lands | How we know |
|---|---|---|
| Speed | Not measured | The first figure we will publish, from logger-car runs on a controlled course. |
| Vehicle length | Not measured | It has to separate a car from a van, and we don't know yet whether it can. |
| Count, one lane | Not measured | We expect to miss motorcycles and vehicles arriving side by side. That is an expectation, not a measurement. |
| Class, car / van / truck | Not measured | We expect small trucks to read as vans. There is no evidence either way yet. |
| Night, no street light | Not measured | LiDAR should carry this on its own, with the camera adding almost nothing. We haven't confirmed it. |
| Count, four lanes | Not measured | A different problem from one lane, and we have run neither. |
| Heavy rain | Not measured | Untested, like every row above it. |
| Snow and plow spray | Not measured | The condition a plow buyer cares most about, and the furthest from an answer. |
| A junction | Not measured | Turning movements and stopped queues are a different problem from a straight mile. |
Prototype validation is the next step
The prototype gives the team hardware to test. Results will replace the not-measured entries after each method and condition is checked.
A controlled course answers limited questions
Closed-course tests isolate sensing and timing. Public-road tests add real traffic, site geometry, installation constraints, and changing conditions.
Failures stay in the record
When a check fails, the run remains in the validation record with the reason and affected fields.