Swift ARKit - how to track moving pet - ios

Question

I need to write an application that can recognize a cat or a dog and measure the number of breaths per minute.
For this, I would like to take 100 colors of points from the cat area every 100 milliseconds. These points will be processed by Fourier Transform function that measures cat's breaths based on cyclical color changes.
The Vision library can detect cat or dog by the camera snapshot, but I'm afraid that every 100ms the ARAnchor will be placed elsewhere and I won't be able to get the same 100 points from the cat area.
Here is my ViewController:
import SceneKit
import ARKit
import Vision
class ViewController: UIViewController, ARSCNViewDelegate, ARSessionDelegate {
var detectionAvailableUntil: Int64 = .zero
#IBOutlet var sceneView: ARSCNView!
private var viewportSize: CGSize!
override func viewDidLoad() {
super.viewDidLoad()
sceneView.delegate = self
viewportSize = sceneView.frame.size
}
override func viewWillAppear(_ animated: Bool) {
super.viewWillAppear(animated)
resetTracking()
}
override func viewWillDisappear(_ animated: Bool) {
super.viewWillDisappear(animated)
sceneView.session.pause()
}
private func resetTracking() {
let configuration = ARWorldTrackingConfiguration()
configuration.planeDetection = []
sceneView.session.run(configuration)
}
lazy var catRequest: VNRecognizeAnimalsRequest = {
do {
let req = VNRecognizeAnimalsRequest() { [weak self] request, error in
self?.processCatDetections(for: request as! VNRecognizeAnimalsRequest, error: error)
}
return req
} catch {
fatalError("Fatal error")
}
}()
func processCatDetections(for request: VNRecognizeAnimalsRequest, error: Error?) {
guard error == nil else {
print("Object detection error: \(error!.localizedDescription)")
return
}
if let result = request.results?.first as? VNRecognizedObjectObservation {
let cats = result.labels.filter({$0.identifier == "Cat"})
for _ in cats {
guard let currentFrame = self.sceneView.session.currentFrame,
result.confidence > 0.3 else { continue }
let fromCameraImageToViewTransform = currentFrame.displayTransform(for: .portrait, viewportSize: self.viewportSize)
let boundingBox = result.boundingBox
let viewNormalizedBoundingBox = boundingBox.applying(fromCameraImageToViewTransform)
let t = CGAffineTransform(scaleX: self.viewportSize.width, y: self.viewportSize.height)
let viewBoundingBox = viewNormalizedBoundingBox.applying(t)
let midPoint = CGPoint(x: viewBoundingBox.midX, y: viewBoundingBox.midY)
let results = self.sceneView.hitTest(midPoint, types: .featurePoint)
guard let result = results.first else { continue }
let anchor = ARAnchor(name: "catAnchor", transform: result.worldTransform)
self.sceneView.session.add(anchor: anchor)
}
}
}
func renderer(_ renderer: SCNSceneRenderer, willRenderScene scene: SCNScene, atTime time: TimeInterval) {
guard let capturedImage = sceneView.session.currentFrame?.capturedImage else { return }
let requestHandler = VNImageRequestHandler(cvPixelBuffer: capturedImage)
do {
try requestHandler.perform([self.catRequest])
} catch {
print("Error: Vision request failed")
}
}
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
let date = Int64(NSDate().timeIntervalSince1970 * 1000)
guard anchor.name == "catAnchor",
date > self.detectionAvailableUntil else { return }
print("catAnchor added")
self.detectionAvailableUntil = Int64(NSDate().timeIntervalSince1970 * 1000 + 500)
let sphereNode = SCNNode(geometry: SCNSphere(radius: 0.01))
sphereNode.geometry?.firstMaterial?.diffuse.contents = UIColor.red
sphereNode.simdWorldTransform = anchor.transform
node.addChildNode(sphereNode)
}
}
The problem is the sphereNode is drawing in different places not on the cat's area (belly or head).
Apple provides libraries for tracking human body or face so I suppose that I need something like ARFaceAnchor or ARBodyAnchor but for cat.
If the app is able to track the cat, it will be able to retrieve the same points within the ARAnchor.
Is it possible?