← Back
Built for my day job

Click-Change Detector: from a screen recording to a finished simulation.

A click-through software simulation lets people practice in a safe copy of real software. Building one means capturing every screen, removing the mouse cursor, and marking every click by hand. On a long recording that’s hours of tedious work, so I built a tool to do it.

53/53 screen changes caught≥96% precision5.9× faster than real time0 bytes sent off the computer

How it works

You drop in a recording. The tool finds each click and screen change, cleans up the frames, and lets you review the result in a browser before exporting it.

1RecordAny screen capture of the software.
2DetectFind every click and what changed on screen.
3CleanRemove the cursor from each frame.
4ReviewKeep, cut or zoom steps in a local web app.
5ExportNumbered steps and click points for Storyline.

Every step gets an honest label, so a reviewer knows exactly where to look:

CleanReconstructedInpainted, flaggedCursor present, flagged

Decisions behind it

No machine learning, on purpose

It uses plain computer vision (OpenCV and NumPy), so the same video always gives the same result, and it’s easy to explain why it made each call.

Nothing leaves the computer

Training recordings can show internal systems. Everything runs locally, and nothing is uploaded anywhere.

Checked against something independent

I tested it against a separate detector on a real recording: it caught all 53 screen changes, with at least 96% precision.

It admits what it can’t do

When it can’t remove a cursor cleanly, it says so and flags the step for a person to check, instead of hiding the problem.

Python · OpenCV · NumPyAbout 8,400 lines, with testsA local web app for reviewVersion 0.7

Want the details?

Portland, Oregon. Open to remote work or relocating.

pattersonjsds@gmail.com