Drone building inspection

PFI Sorting — turning an inspection flight's photo stream into sorted inspection runs

A sorting tool that reads drone photo metadata and splits a building-inspection flight into numbered inspection run folders, using the straight-down camera marker pilots already fly.

Relationship
Client project
Industry
Drone building inspection
Status
Live

The product

Interface summary

What the sorting tool puts in front of an operator

  • A folder plan previewed before anything is written
  • Numbered inspection run folders, one per detected boundary
  • An adjustable pitch tolerance, defaulting to 2°
  • An opt-in altitude-reversal fallback for flights with no marker
  • A CSV report recording why each folder was started
Written summary, not a screenshot. No product image is published for this build yet.

Client project

Built for an external client.

Who uses it

Whoever has to work through the photographs an inspection flight produced.

Before
An inspection flight comes back as one long photo stream, and someone separates hundreds of images into roof faces, elevations and passes by hand.
After
The tool reads the marker already sitting in the photo metadata and writes the flight out as numbered inspection run folders, with a report of why each folder was started.

The problem

The tool runs on drone building inspections, after the flight has landed and the images are back on a machine.

An inspection flight comes back as one long, undifferentiated photo stream, but the review work is organised by roof face, elevation and inspection pass. Pilots already mark those boundaries in the air by pointing the camera straight down at the start of each one, so the split points are sitting in the photo metadata — until something reads them, someone separates hundreds of images by hand.

What PSOI Tech delivered

  • Pitch-marker detection across the embedded metadata fields drone vendors use
  • Automatic split of a flight into numbered inspection run folders
  • Optional altitude-reversal fallback for flights missing a marker
  • Command line, local web interface and browser app over one sorting engine

Decisions that shaped it

  1. 01

    The split signal is a habit pilots already have

    Pilots point the camera straight down to mark the start of a new roof face, elevation or pass, so that gesture was already in the flight data. Reading it meant the tool changed nothing about how a flight is flown — no new field procedure, no markers to remember, and every previously flown set of images sorts on the same rule.

  2. 02

    Both +90° and -90° count as down, within a tolerance

    Vendors disagree on the sign of a pitched-down reading, and a hand-flown marker is never exactly perpendicular. The detector accepts either sign and defaults to a 2° tolerance, so -89.4° and 91.2° both register as markers rather than being silently missed on one operator's airframe.

  3. 03

    The altitude fallback is opt-in and has to be confident

    Flights that missed a marker can be split on sustained altitude reversals instead, but that inference is off by default and stays deliberately conservative: it ignores small changes, requires several confirmed steps and real vertical span either side of a turn, and stands down around genuine pitch markers. A missed split is easy to spot and fix; an invented one quietly corrupts the run boundaries.

  4. 04

    Nothing is destructive and nothing leaves the machine

    Files are copied rather than moved unless the operator asks otherwise, a preview mode shows the folder plan before anything is written, and a CSV report records why each folder was started. The web interface runs locally against the same engine, so inspection imagery is never uploaded anywhere to be sorted.

Where it stands

A working sorting tool for drone building inspections, usable from the command line, from a local web page, or in the browser.

No performance figures are published for this build. PSOI Tech publishes outcome numbers only where the client has measured and approved them, so this entry describes the work rather than claiming a result.

Stack

  • Python
  • Vercel

Related service

This work maps to the AI & Workflow Automation path — One operational workflow automated, with the controls to run it safely.

AI & Workflow Automation: scope, price and timeline

Its published range sits alongside every other one on the pricing page.

More work

All case studies

Next step

Want something like this for your operations?

Describe the operating problem and the deadline. We will tell you whether it is a fit and what the first step should be.