when the Enable Automatic Camera Overlay is enabled you will get the People Occlusion image overlaid above everything so no objects can be in-front of it. And if you disable Automatic Camera Overlay you won't get access to the AR Texture of People Occlusion to assign it to your own material instance and therefore position in depth and have virtual objects both in front and behind.
The desired outcome might be clearer here in Apples ARKit 3 documentation: [Occluding Virtual Content with People | Apple Developer Documentation|https://developer.apple.com/documentation/arkit/camera_lighting_and_effects/occluding_virtual_content_with_people]
You can see how People Segmentation with Depth should allow for people to appear in front of and behind objects depending on their depth from camera.
I'm not sure which method is valid/correct, or if there is a different way to do this.
There's no existing public thread on this issue, so head over to Questions & Answers just mention UE-116444 in the post.