INFRA INTELLIGENCE
JPEN
Contact

AI models, turned into apps for the field

Model Lab's open models, or models made to your needs, built into phone and web apps.

The models are free, but in the field you need a system that spots things in camera video, alerts people and keeps a record. That system is what gets built.

Telling snowplows where they must not throw snow

A phone in the vehicle spots the roadside signs and announces where the zone starts and ends.

No-snow-throwing signs on the left shoulder of an expressway. The near start sign (a blue square inside a yellow square) and the distant end sign (a yellow and blue cross) are in blue boxes. The truck ahead is blurred.Start and end signs along a sound barrier, each in a blue box labeled Start or End. Vehicles and guide signs are blurred.A yellow and blue cross-shaped end sign behind a roadside fence, in a blue box. Vehicles and guide signs are blurred.

The model's boxes, shown as-is on driving footage it was not trained on.

Detects
Signs marking the start and end of the zone
Runs on
A phone in the vehicle
When found
Sounds an alert and records the spot and photo
Runtime and evaluation details
Runs on
On the phone. The same model runs twice: on an unscaled crop of part of the frame for distant signs, and on the scaled-down full frame for near ones.
Evaluation
On 57 test images not used in training, it detected 12 of 18 start signs and 17 of 21 end signs, with no start/end mix-ups. Only signs inside the distant crop were counted. Night and snowy conditions have not been tested yet.
No-snow-throwing sign on a red and white post: a blue square in the center of a yellow square (start)
Start. A blue square inside a yellow square
No-snow-throwing sign on a red and white post: a yellow and blue cross (end)
End. A yellow and blue cross
See this model in Model Lab
FIG. 01
How it works. A model on the in-car phone finds the sign in the forward view, confirms it after seeing it several times on approach, sounds an alert, and records the spot and photo on a map. 01Forward view IN　1280 × 736 px 02Detect on device ON-DEVICE 03Confirm, then alert CONFIRM ／ ALERT 04Record spot and photo LOG ／ MAP Start 2 times ApproachSound Photo
More uses

Map every sign and curve mirror

Just drive, and you get a list of what to inspect.

Record potholes as you drive

Pothole photos and locations build up on the map.

Blur vehicles before publishing

Vehicles in photos and video are found and blurred automatically.

Find rust and peeling in inspection photos

Send photos, and they come back with the spots to check marked.

When it finds one, color and sound tell you

Red and green screens and sounds do the talking, so no one has to touch the screen while driving.

  1. Scanning

    Looking for signs ahead

    StopReset
    DriveRecordsSettings

    01Once moving, it looks for signs in the forward view.

  2. No throwing

    Start sign confirmed

    StopReset
    DriveRecordsSettings

    02On finding the start sign, the screen turns red and a buzzer sounds.

  3. Throwing OK

    Passed the end sign

    StopReset
    DriveRecordsSettings

    03After passing the end sign, it turns green and a chime sounds.

  4. The drive log screen. The route is drawn in blue on a map, with numbered red and green markers where start and end signs were found. Below: a tally of 2 starts, 2 ends and 18 photos, and photos of the signs found.

    04Each spot is recorded with a photo and can be reviewed on a map.

Map: GSI Tiles (Geospatial Information Authority of Japan)

No signal needed
It runs on the phone, so it works in tunnels and mountains.
For other uses
Screens and alerts are designed around who does what after a detection.
  1. 01

    Open models, ready to try

    If a Model Lab model does the job, work starts straight on the app.

  2. 02

    Nothing suitable? Build the model too

    A few hundred photos are enough to start.

  3. 03

    Fit to the people and the place

    Phones, field devices or the cloud, whatever suits the job.

From request to delivery

Deliverables

  • App with AI models built in (phone or web)
  • Admin dashboard for records (optional)
  • AI models used in the app
  • Evaluation report
  1. 01

    Consultation

    Talk through what to detect and where it will be used, and shape the app.

  2. 02

    The model

    Use an open model or build a new one.

  3. 03

    App development

    Alerts and records are designed around how the app will be used.

  4. 04

    Try and refine

    Use it in the field, and the model and app are improved from there.

FAQ

Q.01Are Model Lab's models free?

Yes. The fee is for developing the app.

Q.02Can you detect something no open model covers?

Yes. A model is built for what you need to find.

Q.03How accurate is it?

It depends on the target and the images. You see the results of a small trial before deciding whether to continue.

Q.04Who owns the app?

This is agreed case by case in the contract. Data you provide is never used for other purposes.

Start by describing your situation

Whether it is a MICHIKARTE trial or an AI app, your plans do not need to be final. Describe the photos or video you have and what you are struggling with, and you will get an email reply on what can and cannot be done. The contact form is in Japanese, but you are welcome to write in English.

Get in touch
Contact