My use case is I want to calculate various gestures of a hand (the first hand) seen by the camera. I am able to find body anchors and hand anchors and poses. See my video here.
I am trying to utilize previous position SIMD3 information to calculate what kind of gesture was demonstrated. I did see the example posted by Apple which shows pinching to write virtually, I am not sure that a buffer is the right solution for something like this.
A specific example of what I am trying to do is detect a swipe, long-press, tap as if the user is wearing a pair of AR glasses (made by Apple one day). For clarification I want to raycast from my hand and perform a gesture on an Entity or Anchor.
Here is a snippet for those of you that want to know how to get body anchors:
public func session(_ session: ARSession, didUpdate frame: ARFrame) {
let capturedImage = frame.capturedImage
let imageRequestHandler = VNImageRequestHandler(cvPixelBuffer: capturedImage,
orientation: .right,
options: [:])
let handPoseRequest = VNDetectHumanHandPoseRequest()
//let bodyPoseRequest = VNDetectHumanBodyPoseRequest()
do {
try imageRequestHandler.perform([handPoseRequest])
guard let observation = handPoseRequest.results?.first else {
return
}
// Get points for thumb and index finger.
let thumbPoints = try observation.recognizedPoints(.thumb)
let indexFingerPoints = try observation.recognizedPoints(.indexFinger)
let pinkyFingerPoints = try observation.recognizedPoints(.littleFinger)
let ringFingerPoints = try observation.recognizedPoints(.ringFinger)
let middleFingerPoints = try observation.recognizedPoints(.middleFinger)
self.detectHandPose(handObservations: observation)
} catch {
print("Failed to perform image request.")
}
}
I am using Swift to detect squares on an image and I can't seem to get it to detect them. It seems to detect rectangles sometimes not sure if this is the correct approach. I am new to swift and image detection so if there is something else I can be doing to detect squares I would greatly appreciate getting pointed in the right direction.
From what I have found on searches is issues around detecting squares / rectangles and perspective. Not sure if this is the issue or just my lack of knowledge of Swift and image detection.
Test image
lazy var rectangleDetectionRequest: VNDetectRectanglesRequest = {
let rectDetectRequest = VNDetectRectanglesRequest(completionHandler: self.handleDetectedRectangles)
// Customize & configure the request to detect only certain rectangles.
rectDetectRequest.maximumObservations = 8 // Vision currently supports up to 16.
rectDetectRequest.minimumConfidence = 0.6 // Be confident.
rectDetectRequest.minimumAspectRatio = 0.3 // height / width
return rectDetectRequest
}()
fileprivate func handleDetectedRectangles(request: VNRequest?, error: Error?) {
if let nsError = error as NSError? {
self.presentAlert("Rectangle Detection Error", error: nsError)
return
}
// Since handlers are executing on a background thread, explicitly send draw calls to the main thread.
DispatchQueue.main.async {
guard let drawLayer = self.pathLayer,
let results = request?.results as? [VNRectangleObservation] else {
return
}
self.draw(rectangles: results, onImageWithBounds: drawLayer.bounds)
drawLayer.setNeedsDisplay()
}
}
I have also changed minimumAspectRatio to 1.0 which from what information I have found would be a square and it still did not give the expected results.
Vertical place detection in ARKit is not very good so I am using proximity sensor to place found vertical plane. The UX is as follows:
Ask user to place front of the device on the wall
When the proximity sensor is triggered make a vertical plane using AR Camera transform
The issue that I am currently facing is that when front sensor is triggered, everything comes to a halt. All the CoreMotion sensors and render methods for ARSCNViewDelegate calls stop. This causes the world origin to move from its origin point and make the placed item also move with it.
Is there a way to get proximity sensor data without shutting down everything? Is there a better way to place vertical items?
Asynchronous functions let you perform two or more tasks almost simultaneously, without waiting until the dispatched block is executed.
So, you have to use an approach like this:
class ViewController: UIViewController {
override func viewWillAppear(_ animated: Bool) {
super.viewWillAppear(animated)
DispatchQueue.main.async {
self.proximitySensorActivation()
}
}
func proximitySensorActivation() {
let device = UIDevice.current
device.isProximityMonitoringEnabled = true
if device.isProximityMonitoringEnabled {
NotificationCenter.default.addObserver(self,
selector: #selector(proximityChanged),
name: UIDevice.proximityStateDidChangeNotification,
object: device)
}
}
#objc func proximityChanged(notification: NSNotification) {
if let device = notification.object as? UIDevice {
print(device)
}
}
}
Desired behavior is: when an action is removed from a node (with removeAction(forKey:) for instance) it stops to animate and all the changes caused by action are discarded, so the node returns back to pervious state. In other words, I want to achieve behavior similar to CAAnimation.
But when a SKAction is removed, the node remains changed. It's not good, because to restore it's state I need to know exactly what action was removed. And if I then change the action, I also will need to update the node state restoration.
Update:
The particular purpose is to show possible move in a match-3 game. When I show a move, pieces start pulsating (scale action, repeating forever). And when the user moves I want to stop showing the move, so I remove the action. As the result, pieces may remain downscaled. Later I would like to add more fancy and complicated animations, so I want to be able to edit it easily.
Thanks to the helpful comment and answer I came to my own solution. I think the state machine would be bit too heavy here. Instead I created a wrapper node, which main purpose is run the animation. It also has a state: isAimating property. But, first of all, it allows to keep startAnimating() and stopAnimating() methods close to each other, incapsulated, so it's more difficult to mess up.
class ShowMoveAnimNode: SKNode {
let animKey = "showMove"
var isAnimating: Bool = false {
didSet {
guard oldValue != isAnimating else { return }
if isAnimating {
startAnimating()
} else {
stopAnimating()
}
}
}
private func startAnimating() {
let shortPeriod = 0.2
let scaleDown = SKAction.scale(by: 0.75, duration: shortPeriod)
let seq = SKAction.sequence([scaleDown,
scaleDown.reversed(),
scaleDown,
scaleDown.reversed(),
SKAction.wait(forDuration: shortPeriod * 6)])
let repeated = SKAction.repeatForever(seq)
run(repeated, withKey: animKey)
}
private func stopAnimating() {
removeAction(forKey: animKey)
xScale = 1
yScale = 1
}
}
Usage: just add everything that should be animated to this node. Works well with simple animations, like: fade, scale and move.
As #Knight0fDragon suggested, you would be better off using the GKStateMachine functionality, I will give you an example.
First declare the states of your player/character in your scene
lazy var playerState: GKStateMachine = GKStateMachine(states: [
Idle(scene: self),
Run(scene: self)
])
Then you need to create a class for each of these states, in this example I will show you only the Idle class
import SpriteKit
import GameplayKit
class Idle: GKState {
weak var scene: GameScene?
init(scene: SKScene) {
self.scene = scene as? GameScene
super.init()
}
override func didEnter(from previousState: GKState?) {
//Here you can make changes to your character when it enters this state, for example, change his texture.
}
override func isValidNextState(_ stateClass: AnyClass) -> Bool {
return stateClass is Run.Type //This is pretty obvious by the method name, which states can the character go to from this state.
}
override func update(deltaTime seconds: TimeInterval) {
//Here is the update method for this state, lets say you have a button which controls your character velocity, then you can check if the player go over a certain velocity you make it go to the Run state.
if playerVelocity > 500 { //playerVelocity is just an example of a variable to check the player velocity.
scene?.playerState.enter(Run.self)
}
}
}
Now of course in your scene you need to do two things, first is initialize the character to a certain state or else it will remain stateless, so you can to this in the didMove method.
override func didMove(to view: SKView) {
playerState.enter(Idle.self)
}
And last but no least is make sure the scene update method calls the state update method.
override func update(_ currentTime: TimeInterval) {
playerState.update(deltaTime: currentTime)
}
I would like to check whether the ARReferenceImage is no longer visible in the camera's view. At the moment I can check if the image's node is in the camera's view, but this node is still visible in the camera's view when the ARReferenceImage is covered with another image or when the image is removed.
func renderer(_ renderer: SCNSceneRenderer, updateAtTime time: TimeInterval) {
guard let node = self.currentImageNode else { return }
if let pointOfView = sceneView.pointOfView {
let isVisible = sceneView.isNode(node, insideFrustumOf: pointOfView)
print("Is node visible: \(isVisible)")
}
}
So I need to check if the image is no longer visible instead of the image's node visibility. But I can't find out if this is possible. The first screenshot shows three boxes that are added when the image beneath is found. When the found image is covered (see screenshot 2) I would like to remove the boxes.
I managed to fix the problem! Used a little bit of Maybe1's code and his concept to solving the problem, but in a different way. The following line of code is still used to reactivate the image recognition.
// Delete anchor from the session to reactivate the image recognition
sceneView.session.remove(anchor: anchor)
Let me explain. First we need to add some variables.
// The scnNodeBarn variable will be the node to be added when the barn image is found. Add another scnNode when you have another image.
var scnNodeBarn: SCNNode = SCNNode()
// This variable holds the currently added scnNode (in this case scnNodeBarn when the barn image is found)
var currentNode: SCNNode? = nil
// This variable holds the UUID of the found Image Anchor that is used to add a scnNode
var currentARImageAnchorIdentifier: UUID?
// This variable is used to call a function when there is no new anchor added for 0.6 seconds
var timer: Timer!
The complete code with comments below.
/// - Tag: ARImageAnchor-Visualizing
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
guard let imageAnchor = anchor as? ARImageAnchor else { return }
let referenceImage = imageAnchor.referenceImage
// The following timer fires after 0.6 seconds, but everytime when there found an anchor the timer is stopped.
// So when there is no ARImageAnchor found the timer will be completed and the current scene node will be deleted and the variable will set to nil
DispatchQueue.main.async {
if(self.timer != nil){
self.timer.invalidate()
}
self.timer = Timer.scheduledTimer(timeInterval: 0.6 , target: self, selector: #selector(self.imageLost(_:)), userInfo: nil, repeats: false)
}
// Check if there is found a new image on the basis of the ARImageAnchorIdentifier, when found delete the current scene node and set the variable to nil
if(self.currentARImageAnchorIdentifier != imageAnchor.identifier &&
self.currentARImageAnchorIdentifier != nil
&& self.currentNode != nil){
//found new image
self.currentNode!.removeFromParentNode()
self.currentNode = nil
}
updateQueue.async {
//If currentNode is nil, there is currently no scene node
if(self.currentNode == nil){
switch referenceImage.name {
case "barn":
self.scnNodeBarn.transform = node.transform
self.sceneView.scene.rootNode.addChildNode(self.scnNodeBarn)
self.currentNode = self.scnNodeBarn
default: break
}
}
self.currentARImageAnchorIdentifier = imageAnchor.identifier
// Delete anchor from the session to reactivate the image recognition
self.sceneView.session.remove(anchor: anchor)
}
}
Delete the node when the timer is finished indicating that there was no new ARImageAnchor found.
#objc
func imageLost(_ sender:Timer){
self.currentNode!.removeFromParentNode()
self.currentNode = nil
}
In this way the currently added scnNode wil be deleted when the image is covered or when there is found a new image.
This solution does unfortunately not solve the positioning problem of images because of the following:
ARKit doesn’t track changes to the position or orientation of each detected image.
I don't think this is currently possible.
From the Recognizing Images in an AR Experience documentation:
Design your AR experience to use detected images as a starting point for virtual content.
ARKit doesn’t track changes to the position or orientation of each detected image. If you try to place virtual content that stays attached to a detected image, that content may not appear to stay in place correctly. Instead, use detected images as a frame of reference for starting a dynamic scene.
New Answer for iOS 12.0
ARKit 2.0 and iOS 12 finally adds this feature, either via ARImageTrackingConfiguration or via the ARWorldTrackingConfiguration.detectionImages property that now also tracks the position of the images.
The Apple documentation to ARImageTrackingConfiguration lists advantages of both methods:
With ARImageTrackingConfiguration, ARKit establishes a 3D space not by tracking the motion of the device relative to the world, but solely by detecting and tracking the motion of known 2D images in view of the camera. ARWorldTrackingConfiguration can also detect images, but each configuration has its own strengths:
World tracking has a higher performance cost than image-only tracking, so your session can reliably track more images at once with ARImageTrackingConfiguration.
Image-only tracking lets you anchor virtual content to known images only when those images are in view of the camera. World tracking with image detection lets you use known images to add virtual content to the 3D world, and continues to track the position of that content in world space even after the image is no longer in view.
World tracking works best in a stable, nonmoving environment. You can use image-only tracking to add virtual content to known images in more situations—for example, an advertisement inside a moving subway car.
The correct way to check if an image that you are tracking is not currently tracked by ARKit is by using the "isTracked" property in the ARImageAnchor on the didUpdate node for anchor function.
For that, I use the next struct:
struct TrackedImage {
var name : String
var node : SCNNode?
}
And then an array of that struct with the name of all the images.
var trackedImages : [TrackedImage] = [ TrackedImage(name: "image_1", node: nil) ]
Then in the didAdd node for anchor, set the new content to the scene and also add the node to the corresponding element in the array of trackedImages
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
// Check if the added anchor is a recognized ARImageAnchor
if let imageAnchor = anchor as? ARImageAnchor{
// Get the reference ar image
let referenceImage = imageAnchor.referenceImage
// Create a plane to match the detected image.
let plane = SCNPlane(width: referenceImage.physicalSize.width, height: referenceImage.physicalSize.height)
plane.firstMaterial?.diffuse.contents = UIColor(red: 1, green: 1, blue: 1, alpha: 0.5)
// Create SCNNode from the plane
let planeNode = SCNNode(geometry: plane)
planeNode.eulerAngles.x = -.pi / 2
// Add the plane to the scene.
node.addChildNode(planeNode)
// Add the node to the tracked images
for (index, trackedImage) in trackedImages.enumerated(){
if(trackedImage.name == referenceImage.name){
trackedImage[index].node = planeNode
}
}
}
}
Finally in the didUpdate node for anchor function we search for the anchor name in our array and check if the property isTracked is false.
func renderer(_ renderer: SCNSceneRenderer, didUpdate node: SCNNode, for anchor: ARAnchor) {
var trackedImages : [TrackedImage] = [ TrackedImage(name: "image_1", node: nil) ]
if let imageAnchor = anchor as? ARImageAnchor{
// Search the corresponding node for the ar image anchor
for (index, trackedImage) in trackedImages.enumerated(){
if(trackedImage.name == referenceImage.name){
// Check if track is lost on ar image
if(imageAnchor.isTracked){
// The image is being tracked
trackedImage.node?.isHidden = false // Show or add content
}else{
// The image is lost
trackedImage.node?.isHidden = true // Hide or delete content
}
break
}
}
}
}
This solution works when you want to tracked multiple images at the same time and know when any of them is lost.
Note: For this solution to work the maximumNumberOfTrackedImages in the AR configuration must be set to a nonzero number.
For what its worth, I spent hours trying to figure out how to constantly check for image references. The didUpdate function was the answer. Then you just need to test of the reference image is being tracked using the .isTracked property. At that point, you can set the .isHidden property to true or false. Heres my example:
func renderer(_ renderer: SCNSceneRenderer, didUpdate node: SCNNode, for anchor: ARAnchor) {
let trackedNode = node
if let imageAnchor = anchor as? ARImageAnchor{
if (imageAnchor.isTracked) {
trackedNode.isHidden = false
print("\(trackedNode.name)")
}else {
trackedNode.isHidden = true
//print("\(trackedImageName)")
print("No image in view")
}
}
}
I'm not entirely sure I have understood what your asking (so apologies), but if I have then perhaps this might help...
It seems that for insideOfFrustum to work correctly, that their must be some SCNGeometry associated with the node for it to work (an SCNNode alone will not suffice).
For example if we do something like this in the delegate callback and save the added SCNNode into an array:
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
//1. If Out Target Image Has Been Detected Than Get The Corresponding Anchor
guard let currentImageAnchor = anchor as? ARImageAnchor else { return }
//2. Print The Anchor ID & It's Associated Node
print("""
Anchor With ID Has Been Detected \(currentImageAnchor.identifier)
Associated Node Details = \(node)
""")
//3. Store The Node
imageTargets.append(node)
}
And then use the insideOfFrustum method, 99% of the time it will say that the node is in view even when we know it shouldn't be.
However if we do something like this (whereby we create a transparent marker node e.g. one that has some geometry):
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
//1. If Out Target Image Has Been Detected Than Get The Corresponding Anchor
guard let currentImageAnchor = anchor as? ARImageAnchor else { return }
//2. Print The Anchor ID & It's Associated Node
print("""
Anchor With ID Has Been Detected \(currentImageAnchor.identifier)
Associated Node Details = \(node)
""")
//3. Create A Transpanrent Geometry
node.geometry = SCNSphere(radius: 0.1)
node.geometry?.firstMaterial?.diffuse.contents = UIColor.clear
//3. Store The Node
imageTargets.append(node)
}
And then call the following method, it does detect if the ARReferenceImage is inView:
func renderer(_ renderer: SCNSceneRenderer, updateAtTime time: TimeInterval) {
//1. Get The Current Point Of View
guard let pointOfView = augmentedRealityView.pointOfView else { return }
//2. Loop Through Our Image Target Markers
for addedNode in imageTargets{
if augmentedRealityView.isNode(addedNode, insideFrustumOf: pointOfView){
print("Node Is Visible")
}else{
print("Node Is Not Visible")
}
}
}
In regard to your other point about an SCNNode being occluded by another one, the Apple Docs state that the inViewOfFrostrum:
does not perform occlusion testing. That is, it returns
true if the tested node lies within the specified viewing frustum
regardless of whether that node’s contents are obscured by other
geometry.
Again, apologies if I haven't understood you correctly, but hopefully it might help to some extent...
Update:
Now I fully understand your question, I agree with #orangenkopf that this isn't possible. Since as the docs state:
ARKit doesn’t track changes to the position or orientation of each
detected image.
From the Recognizing Images in an AR Experience documentation:
ARKit adds an image anchor to a session exactly once for each
reference image in the session configuration’s detectionImages array.
If your AR experience adds virtual content to the scene when an image
is detected, that action will by default happen only once. To allow
the user to experience that content again without restarting your app,
call the session’s remove(anchor:) method to remove the corresponding
ARImageAnchor. After the anchor is removed, ARKit will add a new
anchor the next time it detects the image.
So, maybe you can find a workaround for your case:
Let's say we are that structure which saves our ARImageAnchor detected and the virtual content associated:
struct ARImage {
var anchor: ARImageAnchor
var node: SCNNode
}
Then, when the renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) is called, you save the image detected into a temporary list of ARImage:
...
var tmpARImages: [ARImage] = []
func renderer(_ renderer: SCNSceneRenderer, didAdd node: SCNNode, for anchor: ARAnchor) {
guard let imageAnchor = anchor as? ARImageAnchor else { return }
let referenceImage = imageAnchor.referenceImage
// If the ARImage does not exist
if !tmpARImages.contains(where: {$0.anchor.referenceImage.name == referenceImage.name}) {
let virtualContent = SCNNode(...)
node.addChildNode(virtualContent)
tmpARImages.append(ARImage(anchor: imageAnchor, node: virtualContent))
}
// Delete anchor from the session to reactivate the image recognition
sceneView.session.remove(anchor: anchor)
}
If you understood, while your camera's view point out of the image/marker, the delegate function will loop endlessly... (because we removed the anchor from the session).
The idea will be to combine the image recognition loop, the image detected saved into the tmp list and the sceneView.isNode(node, insideFrustumOf: pointOfView) function to determine if the image/marker detected is no longer view.
I hope it was clear...
This code works only if You hold the device strictly horizontally or vertically. If You hold iPhone tilted or starting to tilt if, this code doesn't work:
func renderer(_ renderer: SCNSceneRenderer, updateAtTime time: TimeInterval) {
//1. Get The Current Point Of View
guard let pointOfView = augmentedRealityView.pointOfView else { return }
//2. Loop Through Our Image Target Markers
for addedNode in imageTargets{
if augmentedRealityView.isNode(addedNode, insideFrustumOf: pointOfView){
print("Node Is Visible")
}else{
print("Node Is Not Visible")
}
}
}