← All posts
AI & Edge Computing

Why Truck Parking Counts Go Wrong: Lessons From Iowa and Washington on TPIMS Detection

Julia Skladzinski
November 17, 2026
6
Featured image

A sign that says 12 spaces open when the lot is full is worse than no sign. The driver who trusted it is now in a full lot, out of hours, and looking at the shoulder. Every truck parking information management system lives or dies on whether the count is right, and two state evaluations published in the last three years show exactly where counts go wrong.

This post walks through what Iowa and Washington State learned, what the national design guide says about each detection technology, and what the detection layer has to do to keep the sign honest.

Iowa: the pucks died and nobody noticed for a year

Iowa's share of the eight-state MAASTO system went live in January 2019 with in-pavement sensor pucks in each space at most of its 43 sites, plus radar in/out counters at a few private truck stops. Iowa State University's 2023 evaluation is one of the most candid documents in the field.

The system's accuracy standard was 85 percent for small sites and 90 percent for large ones. Against that standard, the flow data was accurate only 60 percent of the time at small sites and 30 percent of the time at large ones, which the evaluators note "is consistent with driver perceptions." The pucks "started to fail after one and a half years at some sites, but the failures were not discovered until about a year later." A software bug had been overwriting the fields meant to log manual accuracy checks, so the one record that should have caught the problem was blank. When a puck died, its space simply froze in its last state.

The frustrating part is that the system worked when the count was right. Parking spread more evenly along I-80. Downtime stayed under 5 percent. The share of roadside violations that were hours-of-service violations fell from 1.82 percent to 1.08 percent, a 40.7 percent reduction. The evaluators' recommendation for every future deployment: "continuous monitoring of sensor health and independent verification of parking data."

Washington: in/out counts drift until someone resets them

Washington State's March 2026 WSDOT research report focused on the other counting method. Most of its sites use radar in/out counters at the driveways, which are cheaper and scale to any lot size. The report's summary of the weakness is precise: a miscount "will be carried on until a manual check is made or an opposite miscount is made." In a small or medium lot, a handful of misses quickly becomes a significant percentage of capacity.

Comparing sensor output to ground truth from CCTV, the researchers saw two patterns. In the good case, the count tracked reality with small, stable error. In the bad case, the gap "gradually increased" as trucks came and went, until the reported availability "could deviate substantially from the actual parking condition." The causes they documented are the everyday life of a rest area:

  • Trucks bunched at the entrance so closely the counter could not separate them.
  • A trailer dropped across two spaces, or a rig parked with its tail in the next slot, so the lot was fuller than the math said.
  • Trucks parked off-spot on the shoulder of the lot, outside any sensor's view.

They also tested cameras as the independent check and hit the familiar wall. In heavy rain, fog, and snow, "visibility may be reduced, which can degrade detection accuracy," and nighttime, glare, and shadows do the same. Their conclusion was that "radar-based sensors tend to be more robust in poor lighting and weather conditions," cameras add visual confirmation, and combining the two is what improves reliability.

What the national guide says about each sensor

The 2025 NCHRP Guide for Truck Parking Information Management Systems catalogs every detection option in use. The pattern across them is that each one has a failure mode that costs operator time.

  • In-pavement magnetometer pucks. "Reliable and inexpensive," but battery life "has been the most significant challenge identified." They must be replaced when the lot is resurfaced, rebar in concrete can interfere, and freezing climates can affect them.
  • Video at the entrance. "Video cameras on their own do not provide accurate counts." With strong analytics they can, but weather and lighting hamper them.
  • Radar or laser in/out counters. Both "can have in/out counting errors and need frequent updates." One Wisconsin radar unit "registered a zero-turn lawnmower as a truck." Laser is the most expensive entrance option.
  • Per-space video. Highly accurate, but each lot "requires detailed setup and calibration" in all weather conditions, and larger lots need more cameras.

The operational consequence is spelled out: "accuracy checks and count resets will be required once or twice a day in most cases." Indiana recalibrated daily. Tennessee stopped using its system because it "was cumbersome to keep calibrated." Michigan "supplemented off-the-shelf devices with human error correction." Washington's own review found that most technologies can exceed 95 percent accuracy "when properly calibrated," and that qualifier is where the budget goes.

What the detection layer has to do

Put the three sources together and the requirements for a count that stays right write themselves.

  1. Classify at the entrance. A rest area driveway sees cars, pickups, RVs, and bobtails. A counter that cannot tell a tractor-trailer from a pickup is adding cars to the truck count and subtracting them when they leave. Classification is not a feature. It is the count.
  2. Separate vehicles that arrive together. Washington's biggest error source was bunching at the entrance. A sensor has to track individual objects in three dimensions, with their own speed and range, so two trucks nose to tail are two trucks.
  3. Work at night and in weather, with proof. Most parking happens in the dark, and the Jason's Law survey puts unauthorized parking squarely between 7 p.m. and 9 a.m. The sensor has to perform in exactly the conditions cameras lose, and the spec should require evidence.
  4. Verify itself. Iowa's lesson is that a sensor nobody is watching will fail silently. The detection layer should report its own health, keep a visual record an operator can check remotely, and flag when the count stops making sense.
  5. Cross-check in/out against occupancy. The strongest configuration is a count at the entrance and a view of the spaces from the same device, so drift shows up as a discrepancy instead of a complaint.

Fusion at the entrance

FusionSensor was built in a Colorado truck parking pilot for exactly this list. HD3D radar tracks every vehicle in three dimensions, so bunched trucks resolve as separate objects with their own speed and range, in snow, fog, and darkness. HD video, fused with the radar 20 times a second on the device by TrueEdge, supplies the classification: tractor-trailer, bobtail, pickup, car, RV, at 98.7 percent accuracy. The same unit at a lot entrance does the in/out count and can watch spot-specific occupancy as the cross-check, reports its health through Omnisight Connect, and keeps clips an operator can pull up when a number looks off. Washington's 2026 report reviewed Omnisight's radar and video fusion sensor among current detection options and cited detection accuracy of approximately 98 percent.

It installs on an existing pole in under an hour with no pavement cut and no batteries to die, which is the part Iowa would have appreciated. See how it runs a corridor on our truck parking management page, read how a TPIMS works end to end, or use our specification checklist to put these requirements in writing.

Want more like this?

Subscribe to The Edge Brief for monthly field notes, deployment stories, and product updates.

Subscribe →