How to Run a Vehicle Classification Count That Holds Up: A Checklist for DOTs and Consultants

Most classification counts that get rejected were not wrong because of the math. They were wrong because of a setup decision made before the sensor was turned on. This checklist covers the choices that decide whether a 13-class count survives review, in the order you make them.

New to the scheme? Read what the FHWA 13 classes are first.

1. Decide what the count is for

A pavement design input, an HPMS sample section, a truck-route study, and a weigh station siting study all need classification data, but they need different things from it. Pavement design cares most about Classes 5 through 13 and their directional split. HPMS cares about meeting the Traffic Monitoring Guide's sampling and duration rules. A truck-route study cares about peak-hour truck percentage by direction. Write the purpose down before picking sites, because it determines everything below.

2. Pick sites the way the guide expects

• Mid-block, away from intersections, driveways, and merge areas, so vehicles are at steady speed in a single lane.
• Straight and level enough that trucks are not shifting lanes or braking through the site.
• One sensor per direction unless the technology covers both reliably.
• Not in a work zone, not during a detour, and not in the week of a local event that distorts the traffic mix.

3. Count long enough, on the right days

The FHWA Traffic Monitoring Guide's baseline for a short-duration classification count is 48 continuous hours, Tuesday through Thursday, outside holiday weeks. Truck traffic is different on Mondays and Fridays and different again on weekends. If the study calls for a weekly profile, count seven days. If the purpose is annual estimates, the count will be factored against a continuous station, so it has to follow the same day-of-week conventions that station uses.

4. Choose a sensor that can see axles in the conditions you will have

The scheme is axle-based. Length-only sensors cannot separate the truck classes that pavement design depends on. Weather and darkness will happen during a 48-hour count, so the sensor has to keep classifying through both. Tubes count axles but fail on high-volume, multi-lane roads and in wet weather. Camera-only loses axles at night. Radar and video fusion reads length and speed from the radar and axles from the video, in any conditions, which is why it has become the default for programs that cannot afford a redo. Our sensor comparison covers each option in detail.

5. Mount for the lanes, not the pole

• Height and offset set to the manufacturer's spec for the number of lanes, not whatever the nearest pole happens to give you.
• Clear line of sight to every lane being counted, with no sign faces or vegetation in the field of view.
• Aim and confirm with a live view before leaving the site. A five-minute check here saves the whole count.

6. Calibrate on site, and record it

Run a short manual check against the sensor's classification for 15 to 30 minutes. Confirm lane assignment, direction, and that Class 9s are landing in Class 9. Log the check with time and observer. Reviewers ask for it, and if a question comes up later, the log is what settles it.

7. Watch the count while it runs

A sensor that reports live lets you catch a problem on day one instead of discovering it on pickup. Look for lanes that drop to zero, directional splits that flip, or a class distribution that suddenly changes. Any of those usually means a knocked tube, a shifted mount, or a lane closure you were not told about.

8. QA before you deliver

• Compare volume and truck percentage to the nearest continuous count station or the last count at the site.
• Check for gaps and flag any hour with missing data rather than silently interpolating.
• Confirm the class distribution is plausible for the route. A rural interstate with no Class 9s is a sensor problem, not a finding.
• Keep the video or event record where the sensor provides one. It is the fastest way to resolve a challenge.

9. Deliver in the format the client already uses

Classification data ends up in ATMS platforms, HPMS submittals, and pavement design software. Delivering in a format that needs re-keying invites errors. Ask for outputs in the formats the agency already consumes, such as NTCIP or MS2, and provide the raw per-vehicle record alongside the hourly summary.

The mistakes that get counts redone

• Counting Friday to Sunday and calling it typical.
• Siting at a signal or ramp where trucks are accelerating or changing lanes.
• A length-based sensor on a study that needs the full 13 classes.
• No calibration log.
• A weather gap in the middle of the 48 hours with no backup.

Every one of those is avoidable at setup. For programs that want to remove the redo risk entirely, FusionSensor classifies into all 13 bins at 98 percent or better, in any weather, with a video record, from one device that goes up once. See how it fits a study program on our traffic studies page, or read common questions about sensor accuracy.