What AI can do with a point cloud, once it can read it

Measure anything

Ceiling height, door width, the area of a room, the angle of a ramp. The AI fits planes and measures regions in the points, then proposes a measurement you can see in the viewer and confirm. Numbers come from the points, never from a picture.

• "How high is the ceiling above the floor in this room, and is the floor level across the whole room?"
• "What is the clear width of the corridor at its narrowest point?"

Check floors, ceilings and walls

Flatness, level and plumb are the questions a scan answers best. The AI lays virtual straightedges over a surface and grades the gaps against DIN 18202, BS 8204, EN 13670 or ACI 117, estimates ASTM E1155 F-numbers and TR34 properties, maps the worst high and low patches and proposes an issue at each. The same check is a tool in the app, included with the AI step and usable without connecting an AI: see floor flatness on iPad.

• "Check the floor slab in the scan I have open against ±10 mm. Pin every area that is out of tolerance as an issue and tell me the worst high and low spot."
• "Is the wall on the left flat enough for tiles under DIN 18202 row 6?"

Count and find

The AI splits a region into ground, planes and connected objects, each with its size, orientation, height above ground, colour, intensity and class, then counts and names what matches your question and highlights it in the viewer. Columns in a hall, pallets in an aisle, pipes under a ceiling, the one cabinet that is missing. Segmentation is geometric rather than a trained detector, so the AI tells you when a shape is ambiguous.

• "Count the columns in this hall and tell me which ones are out of plumb by more than 10 mm."
• "Find anything lower than 2.1 m above the floor in the corridor and pin it as a headroom issue."

Detect deviations

Where two scan stations disagree on the same wall, where a surface leaves tolerance, where the scan has holes. The AI compares stations, grades surfaces and lists thin coverage nearest first, so you know where to put another station before you leave site rather than after you get back.

• "Compare the scan stations in this project. Do any disagree by more than 5 mm on the same walls?"
• "I'm still on site. List the places where coverage is thin or missing, nearest first."

Classify and clean up

Ground points on a roof, vegetation tagged as building, a neighbour's yard you want to hide. The AI selects points by geometry, class, colour or intensity and proposes a class for each group, up to fifty at a time. The file keeps its own classes; the edits apply over them only once you confirm, and you can withdraw any of them.

• "Some roof points are classified as ground. Find them and propose moving them to building."
• "Trim this scan to the site boundary. Everything outside the fence goes into its own class so I can hide it."

Show the client

The AI can move the camera, crop the view and fly a path through the scan on your iPad while you talk. Put the AI app beside Pointscape in Split View and ask for a tour.

• "Fly the camera around the building once, then walk me through the ground floor starting at the entrance."

Write the report

Every issue and measurement the AI proposes waits in Pointscape until you confirm it. Confirmed ones sit in the project beside anything you recorded by hand and appear in the branded PDF report with their snapshots. Ask on site, confirm in the car park, send the PDF before the office opens.

What it will not do

• Guess from a picture. Every number comes from a measurement tool on the points.
• Recognise objects the way a trained detector does. It reasons from geometry and tells you when it is unsure.
• Certify a floor. A flatness result is measured from a scan by the standard's method, and the report says so.
• Edit your project on its own. Proposals only; you confirm.
• Read a scan on your device while Pointscape is in the background. Cloud scans stay reachable.