Image detection accuracy - even higher!

How can I improve further the accuracy of the image detected?

I have tested with the matching threshold at the minimum and maximum, but there seems to be another issue.

The fact is that the actions simply pick the “most similar” spot on the screen. You can add some noise to an image and still the robot will do the task. I would like the inaccuracy to trigger an error that the robot should deal with.

For example: if there is a red mark within the anchor area on your target image, but not in your screen any longer, then the robot should not carry on simply ignoring the unmatched red mark.

@Guillermo_Lasso5,

Please post here image examples (with and without red dot)

These three images work all the same:

1 Like

Thanks for sharing this case. Currently the image similarity algorithm does not take into account solid colors. Please vote for this feature to raise its priority.

Thanks Sasha. Here is a more critical, but actual example… Especially for those considering to vote:

Both are seen as equal by the robot, using the maximum image matching threshold.

1 Like

That’s interesting… thanks for sharing this. Wondering what’s the algorithm behind image recognition

Hi,

I tested RPA on recognising colours - basically, i created a red rectangle and green rectangle. RPA is supposed to image recognise the colours. Eg if it recognises the red rectangle, it is supposed to write down ‘red’ by using booleans and if-else condition.

However, RPA will always fail the test and the boolean will remain false. However, when i ask it to recognise an area, which includes the red/green rectangle, it is able to do so. Is this a bug? That is cannot recognise colour purely? This colours i used are not found elsewhere on the screen at all. Does not work on blue colour also.

Also, does anyone face the issue of something like below popping out?

this happen randomly, but whenever i manually recapture the image to be recognised by the mouse clicking. I can make it go away by reclicking on the step but it’s quite irritating as at times i forget to do so and RPA will not work when going to the next step (because i have no idea is it trying to switch windows or what not).

Thanks

please see the posts above

Hi @azinchuk

Thank you !

What about the second issue? Has anyone encountered this before?

please create a video or a gif file of the problem because it is hard to understand what issue you are experiencing.

We are now working on improving image-recognition.
Feel free to add the problems you currently have with image-based actions in this topic