3 Replies
      Latest reply on Jun 13, 2019 1:01 AM by atmorales
      JasonRTBond Level 1 Level 1 (0 points)

        I've been checking out the ARKit 3 beta examples and am specifically interested in tracking a person's body while ALSO using person segmentation. (Use case: a person moves among virtual objects, and those objects react to approximated collisions with the user's body. Body Detection used to generate approximate collision for the user's body, and Person Segmentation used to enforce the sense of the user moving between objects.)


        However, my attempts so far to create an AR configuration which allows this have not been successful. Here is just creating a frame semantics object with the desired features:


          var semantics = ARConfiguration.FrameSemantics()
          guard ARWorldTrackingConfiguration.supportsFrameSemantics(semantics) else {
               fatalError("Desired semantics not available.")


        This example throws the indicated error. But if either body detection OR person segmentation is left out, it does not.


        Similarly, this returns false:




        • Am I just doing this wrong? (v possible as I'm new to Swift and native iOS in general — usually use Unity & C#)
        • Or, is this simply not supported? And if so, is this likely to change by time of release?
        • Is there a document somehwere which indicates clearly which ARKit features can be used in conjunction with one another? (I could not find one.)
        • Re: Body Detection & Person Segmentation Simultaneously?
          JasonRTBond Level 1 Level 1 (0 points)

          After a little more checking I see that FrameSemantics is only about 2D information related to the frame. So probably what I want is to create an ARBodyTrackingConfiguration and then add the .personSegmentationWithDepth semantic.


          However, this appears to be unsupported. Can anyone confirm or provide more info on this?

            • Re: Body Detection & Person Segmentation Simultaneously?
              rex Level 1 Level 1 (0 points)

              `I'd like to know more as well.


              I'm still in early prototyping/learning mode, and I wanted to use both .personSegmentationWithDepth & .bodyDetection.


              This doesn't work since the configuration types are different:






              I think you know this, but FYI the PeopleOcclusion demo on the Apple developer site uses .personSegmentationWithDepth.

              • Re: Body Detection & Person Segmentation Simultaneously?
                atmorales Level 1 Level 1 (0 points)

                I'd like to know more as well.


                ARBodyTrackingConfiguration and .personSegmentationWithDepth seem like natural complements in that the combination would enable lots of interesting interactions – but I am also seeing that ARBodyTrackingConfiguration (presently) does not support neither .personSegmentationWithDepth nor .personSegmentation.


                Is this expected behavior?  Any chance this will change in an upcoming beta?