Edge detection - a new way to perform image-based UI testing
03-01, 17:30–17:50 (UTC), Gather Town

My personal "rabbit hole" with image-based UI testing - and how I got back.


"Image-based UI testing libraries (e.g. ImageHorizon or Sikuli) use small reference images to localize elements on the screen (e.g. buttons).

This method works extremely well ... - well, almost.

""Image 'button1.png' was not found on screen"".
Do you know this message? If you still get it after 30 test runs, you want to scream at the library: ""Are you blind? There's the button! Click on it now, for God's sake!""

In a customer project I had such errors all the time - and got to the bottom of them.
And to be honest, I was a bit shocked how lucky I always was when I finally got a hit and the test worked - just because I had fiddled long enough with the recognition accuracy (""similarity""/""confidence"" etc.) in an almost criminal way.

With this in mind, we added a new detection strategy to the ImageHorizonLibrary which pre-processes the images to be searched using the ""canny edge detection"" algorithm.

In the presentation, I introduce the ""edge"" strategy and use examples from real customer projects to show where it has been a real problem solver."

Simon Meggle lives near Augsburg (DE). Since his first contact with the tool "Nagios" in early 2000, the topic "monitoring" has accompanied him his whole professional life: as an IT administrator in various industries, then as a consultant from 2011. With the foundation of his company ELABIT GmbH in 2018, he has finally specialized on the monitoring tool "Checkmk".
During the 2020 pandemic, he developed the Checkmk extension "Robotmk" to integrate Robot Framework, which gained a lot of attention in the monitoring community. 
Since 2022, he works among others for Checkmk as "Principal Product Manager" in the Synthetic Monitoring area to establish Synthetic Monitoring in the monitoring world.