We have noticed that Wikitude Cloud Recognition has 2 steps; the first is recognizing response from the cloud which will generate the marker locally which leads to Wikitude using that marker for local image tracking.
We have encountered a rare problem when we were only able to complete step 1. Sometimes we got recognition response from the cloud server but the model was not tracked properly in step 2.
My question is, is there any workaround for this problem or a way to prevent such a thing from happening? We only got this problem on one of our markers.
could you please provide further details:
Thx and greetings
SDK, Professional Edition 9.4
Happened in both ios and android(Test in Iphone 11,X,8,Android 9 (huaweii y9 2019))
The model was not tracking properly because we have a bad marker(Lots of text). The problem is if the mark is bad, cloud recognition should NOT recognize it anyway. Right now, We got OnRecognitionResponse from CloudTracker, but not the OnImageRecognized from Image Trackable resulting in the model going out of place.
your observation is partly correct. The cloud recognition service is getting a downscaled image for recognition and returns the marker for the recognized target.
Text is always bad for recognition and may/does have an impact on recognition, as you already encountered.
Since the returned tracker works in conjunction with a much higher resolution camera image (where the text's feature points have negative effects on the recognition) the target is not recognized...
There are options to use a SD camera image in the Wikitude Camera (Device Camera on Expert Edition) and disable high definition recognition (= Range Extension to OFF in the Image Tracker)...
But that might not be an option for just one image target.
The best way to tackle this issue would be to try to enhance the quality of the image target itself (see this link for tips & tricks).