能否借助CATransform3D通过ARKit人脸网格获取眼部尺寸？

Hey there! Let's clear up what's going on here and get you on the right track to calculate eye width and interocular distance with ARKit.

First: Why Your Current Values Are Useless #

The data you're getting right now is uninitialized garbage values—here's why:

• ARFaceAnchor.blendShapes[.eyeBlinkLeft] returns a Float (a weight between 0 and 1) that represents how closed the left eye is, not a transform.
• Calling caTransform3DValue on this float is a mistake—this property isn't meant for blend shape weights, so you're just reading random memory values that don't relate to any actual facial measurements.

How to Actually Calculate Eye Width & Interocular Distance #

To get real facial dimensions, you need to use the 3D face mesh vertices from ARFaceAnchor or ARFaceGeometry. Here's how to do it:

Step 1: Understand the Data You Need

• ARFaceAnchor.geometry gives you the full 3D mesh of the face, with vertices in the anchor's local coordinate space.
• You can also use pre-defined facial feature locations from ARFaceAnchor, like leftEye and rightEye (these are SIMD3<Float> positions), or reference specific vertex indices for eye edges (ARKit has standard vertex indices for facial features—you can test by logging vertex positions to identify which correspond to eye edges).

Step 2: Convert Vertices to World Space & Calculate Distances

If you want measurements in real-world units (meters), convert the local vertices to world space using the face node's transform, then compute the distance between relevant points:

func renderer(_ renderer: SCNSceneRenderer, didUpdate node: SCNNode, for anchor: ARAnchor) {
    guard let faceAnchor = anchor as? ARFaceAnchor else { return }
    
    // Calculate interocular distance (distance between eye centers)
    let leftEyeLocalPos = faceAnchor.leftEye
    let rightEyeLocalPos = faceAnchor.rightEye
    
    // Convert local positions to world space
    let leftEyeWorldPos = simd_mul(node.simdTransform, simd_float4(leftEyeLocalPos, 1)).xyz
    let rightEyeWorldPos = simd_mul(node.simdTransform, simd_float4(rightEyeLocalPos, 1)).xyz
    
    let interocularDistance = simd_distance(leftEyeWorldPos, rightEyeWorldPos)
    print("Interocular distance: \(interocularDistance) meters")
    
    // Calculate left eye width (using mesh vertices)
    guard let device = MTLCreateSystemDefaultDevice() else { return }
    let faceGeometry = ARFaceGeometry(device: device)!
    faceGeometry.update(from: faceAnchor.geometry)
    
    // Replace with actual vertex indices for left eye outer/inner edges
    let leftEyeOuterVertex = faceGeometry.vertices[123] 
    let leftEyeInnerVertex = faceGeometry.vertices[456]
    
    let leftEyeOuterWorld = simd_mul(node.simdTransform, simd_float4(leftEyeOuterVertex, 1)).xyz
    let leftEyeInnerWorld = simd_mul(node.simdTransform, simd_float4(leftEyeInnerVertex, 1)).xyz
    
    let leftEyeWidth = simd_distance(leftEyeOuterWorld, leftEyeInnerWorld)
    print("Left eye width: \(leftEyeWidth) meters")
}

Step 3: Notes on Vertex Indices

ARKit's ARFaceGeometry uses consistent vertex indices for key facial features. You can find the exact indices by testing: log a range of vertex positions, then move your eye to see which coordinates shift to match the eye edges.

Key Takeaways #

• Forget the caTransform3DValue from blend shapes—it's completely irrelevant to facial measurements.
• Use ARFaceAnchor's face mesh vertices or pre-defined feature positions to calculate real-world distances.
• All measurements will be in meters (since ARKit uses a real-world coordinate system).

内容的提问来源于stack exchange，提问作者Rohan