frameSemantics

A frame semantic represents 2D information that ARKit extracts from a frame. Set this property to tell ARKit to provide information about one or more ARConfiguration.FrameSemantics every frame.

To get information about the 2D location of a person that ARKit recognizes in a frame, you enable the bodyDetection frame semantic.

if let config = mySession.configuration as? ARBodyTrackingConfiguration {
    config.frameSemantics.insert(.bodyDetection)
    // Run the configuration to effect a frame semantics change.
    mySession.run(config)
}

To indicate that a person should overlap your app's virtual content regardless of the person's depth in the camera's field of view, you enable the personSegmentation frame semantic.

if let config = mySession.configuration as? ARWorldTrackingConfiguration {
    config.frameSemantics.insert(.personSegmentation)
    // Run the configuration to effect a frame semantics change.
    mySession.run(config)
}

This option works for standard renderers like ARView and ARSCNView, and is suitable for virtual reality or green screen scenarios.

To indicate that a person should overlap your app's virtual content only when the person is closer to the camera than the virtual content, you enable the personSegmentationWithDepth frame semantic.

if let config = mySession.configuration as? ARWorldTrackingConfiguration {
    config.frameSemantics.insert(.personSegmentationWithDepth)
    // Run the configuration to effect a frame semantics change.
    mySession.run(config)
}

This operation is done pixel by pixel, according to the renderer's z-buffer, when you use standard renderers like ARView, or ARSCNView.

If you implement your own renderer, then you use the segmentationBuffer and estimatedDepthData to implement people occlusion. See ARMatteGenerator for more information.