Skip to content

Latest commit

 

History

History
84 lines (60 loc) · 5.46 KB

README.md

File metadata and controls

84 lines (60 loc) · 5.46 KB

DepthCamera

DepthCamera

Add To Your Project In your Podfile add pod 'DepthCamera'.

Then, wherever you need to make calls, import DepthCamera into that specific file.

Details About DepthCamera

Usage: Record a video or image in a 2D or 3D format. How: Reads data buffers in using a AVCaptureDataOutputSynchronizerDelegate. If using the 3D environment, it then applies a filter from the depth data. This is the simulated 3D affect. If recording a video, it will asynchronously add images to a video. To achieve this affect, the images are added at an unknown rate. To further explain this, look @ Video Creation Details below. If an image, then it simply records the image. From there, the image or video can be obtained and represented in either a 2D (AVPlayer) format or 3D (ARKit) format. To see how to use each of these, please reference the Video Viewing below. On how to record a video, please reference Video Recording below. Assumptions: The user has a phone that is capable of using these assets.

Current Issues

The color of green used in the ChromaKey is a tad off. Right now it blocks out pretty much any level of green. The end goal is to block out a very specific green.

Video Creation Details

The MTKViewDelegate inside "RealTimeDepthViewController" is what calls the asynchronous calls to adding images to the VideoCreator that creates the video. Well, although, the delegate is supposed to record at a specific FPS, the code inside causes it to record "when it can". This leads to unidentified times per frame. Since that is the case, I have to set a presentation time for each frame. But, those frames can range at unknown times. For my tests, I have seen between 0.04 and 0.08 frame times, which is between 14 - 25 fps. Although it isn't perfect, it's pretty good looking. However, the audio is recorded at a normal rate and can be heard as if it was recorded at a normal rate.

To create a 3D video, we needed a way to distinguish between videos. So, for each image during video creation, we take the depth map and turn all the pixels that are too far away to a green color. We now have a video with a green background. From here, we can filter out the green pixels. The current filtering isn't perfect as it filters out just about every green, but I am working on it. Once filtered using a ChromaKey, we now have our original video.

Video Recording

To take an image:

let controller = RealtimeDepthMaskViewController.createRealTimeDepthCameraVC(imageOrVideoCaptureMode: .photo, completionHandler: { (image, _) in
                    //Do something with image
                }, backgroundImages: nil)
self.present(controller, animated: true, completion: nil)

To take a video:

let controller = RealtimeDepthMaskViewController.createRealTimeDepthCameraVC(imageOrVideoCaptureMode: .video, completionHandler: { (_, url) in
                    //Do something with url
                }, backgroundImages: nil)
self.present(controller, animated: true, completion: nil)

It's as simple as creating a RealTimeDepthViewController and presenting it. You just need to state whether it is a .video or a .photo. From there, just make sure to use the completion handler to access the URL or the UIImage once you are done.

Video Viewing

Note: When I say 2D and 3D, I am talking about the type of viewer. 2D = AVPlayer and 3D = SCNNode / ARKit. I assume the video is a 3D video. If the video taken is not 3D, then it is a normal video and does not need any filtering done to it.

2D Video:

Since the video has a green background, we need to remove it. To do this, we can simply apply a filter to this! The filter is already made, and I plan on creating a method that creates the composition, but have not done so quite yet. From there, we can apply the AVVideoComposition to the AVPlayerItem and we have a normal video!

//You need to create an AVVideoComposition that removes the green pixels using a ChromaKey
let composition = AVMutableVideoComposition(asset: asset) { (request) in
    let source = request.sourceImage.clampedToExtent()
    let filteredImage = RealtimeDepthMaskViewController.filteredImage(ciimage: source)!.clamped(to: source.extent)
    request.finish(with: filteredImage, context: nil)
}

let url = URL(string: "Wherever you store the video")
let asset = AVURLAsset(url: url)
let playerItem = AVPlayerItem(asset: asset)
playerItem.videoComposition = composition

2D Image:

Just get the image of wherever you stored. It should be a black background since an AVPlayer cannot see alpha channel, although the alpha channel for specific pixels is 0.0. Thus, a black background.

3D Video:

The code below speaks for itself. Have a scene with your contents. RealtimeDepthMaskViewController.get3DChromaKey is of type SCNMaterial. From here, we can apply the contents of the scene of your video onto the new material. Then you have your SCNNode that you represent materials onto and apply the ChromaKey onto the node.

let scene = SKScene(size: videoNode.size)
//Add videoNode to scene
scene.addChild(videoNode)

let chromaKeyMaterial = RealtimeDepthMaskViewController.get3DChromaKey()
chromaKeyMaterial.diffuse.contents = scene
node.geometry!.materials = [chromaKeyMaterial]

3D Image:

Just get the image of wherever you stored. It should be a clear background since in ARKit the SCNNode can see alpha channels. The reason this isn't an issue is because an image can contain alpha channels whereas a video cannot.