Dear Wikitude team,
Dear wikitude community,
just getting started with new 7.2 and was wondering about a specific scenario. Object tracking on a specific small object (like with the fire truck) works fine (retrieving the wto file for use in Unity and deploying on Android Galaxy S6/S7/S8 for trials).
Now we'd like to augment a model landscape of approx. 2x1 meters (comparable to the attached photo) and ideally scan the whole landscape before to create a WTO reference file. Later on, we want to overlay some virtual houses on the model landscape. The overlays need to work "out of the box" without user placement during initialization (like "tap to place").
Recording the whole landscape with a single video with lots of different angles won't fit into <100 MB video file to use in wikitude studio to create a wto, I guess. Our first trials do not deliver enough features.
Until now, I've not seen anything similar in the forum (our applied demos). Could you or someone give us a hint how or if we could get this scenario to work?
With the freshly included ARKit/Core maybe we could use a combination of algorithms, like a reference marker to start from and then relying on SLAM/Kit/Core?
Anyone tried a similar scenario?
thanks for the inquiry and the description of the use-case. You are right, that object recognition would currently work not very well on this scenario. Depending on your time-frame there are two solutions I can think of:
- Extended Image Recognition
- Object Recognition 2.0
When using extended Image Recognition your user would need to start at a pre-defined image and then can move (more or less) freely over the scene. The initial image give you a consistent origin for placing your augmentations. In case your users are loosing tracking, they would need to start over at the image.
Object Recognition 2.0 is a feature we are currently working on, which will be fully available with the next release. This feature will cover scenarios like the one you described. In case you are interested drop us an email to email@example.com and we can give you some more insights.
thanks for the specific reply and your info. We will check how to apply extended image recognition to our scenario. Regarding the object recon 2.0, I will def. drop you a mail! Sounds very promising.
Thanks a lot,
just check once again - extended Image Recognition DOES do re-localization, meaning users don't have to start over at the image.
Looking forward to your mail