Technical overview

How {{PRODUCT_NAME}} measures a road.

Name: {{PRODUCT_NAME}}
Brand: {{COMPANY_NAME}}

The detail behind the four steps: what the survey unit carries, how data moves from the car to the public map, and how depth becomes severity.

Data pipeline

Video stays in the car. Faces and number plates are blurred before upload; only records and evidence frames leave the vehicle.

Role	Component	What it does · vendor figures
AI computer	NVIDIA Jetson Orin Nano	Detection, tracking and depth measurement in the car; local SSD; works offline
Front depth camera	Intel RealSense D455	Active IR stereo + RGB · 95 mm baseline · global-shutter depth · 87° × 58° depth FOV · ideal 0.6–6 m · < 2 % error at 4 m · USB-C 3.1
Side cameras	Left + right colour cameras	Kerb-side lanes on both sides in one pass
Motion	Accelerometer + gyroscope	Jolt when a wheel hits a pothole — a second, physical confirmation
Position	Dual-band GNSS receiver	Lane-level location for every detection
Review	10-inch in-car display	Operator sign-in, assigned route, live confirm / reject
Mount	Roof-rack sensor bar	Clamps to a standard rack; no drilling

01
Aligned capture
Depth and colour frames from the D455 are captured together, so every detection box has depth behind it.
02
Fit the road plane
The surrounding road surface is fitted as a plane; depth is measured as distance below that plane, not from the camera.
03
Measure
Maximum depth, width and area of the region below the plane. Values come from a single calibrated sensor on every car.
04
Cross-check
A detection with no depth below the road (a shadow or tar patch) is flagged "flat" and sent to review, not counted.
05
Water
Standing water reads shallow. Water-filled potholes are flagged "water suspected" and are never reported as shallow.