I am new to AR and trying to develop a web-based augmented reality app for to place large 3D objects in the real world. I need SLAM features such as being able to anchor the object in 3D space in a fixed location, so it responds to the movements/rotations of the view and device accordingly.
However, I was told that the object is supposed to blend in to its environment, meaning it should "respond" by not colliding with them or rendering over real objects that are in front of it - likewise not show real objects that are behind it in space. I believe this concept is called "occlusion". As far as I understand, this concept is computationally quite intense, since it needs to actually understand the environment and detect surfaces etc.
The situation is the following: It is supposed to run on an iPad Air 5th Generation. I understand that a new browser API standard called WebXR is being developed and implemented to various degrees by different browsers, which is supposed to make such things easier. However, Apple has still not released WebXR for mobile Safari as of today (and all other browsers on mobile iOS use the Safari engine). So i can't use WebXR. Furthermore, the iPad Air also does not have a LiDAR Scanner for enhanced depth detection.
I read that there are techniques for SLAM with surface detection based on approximating depths with RGB from the camera etc., but they tend to not work so well and the performance is likely to be unusble for web browsers (although the device itself has pretty good hardware).
Is SLAM like this possible in mobile safari? Even if currently not widely used or available, is it theoretically possible? Even if i would have to implement the calculations of depth perception etc. all by myself in WebAssembly for example.