Overview

In today's virtual conferencing landscape, the ability to seamlessly swap out your surroundings with personalized images or videos is not just a feature, but a necessity for fostering a professional and engaging remote presence.

👉 By end of this wiki, you can expect the virtual background feature to look like this:

Warning

In this document, I use VisionKit to separate background, but VNGeneratePersonInstanceMaskRequest only support iOS 17+

Common WebRTC terms you should know

1. VideoFrame: It contains the buffer of the frame captured by the camera device in I420 format.
2. VideoSink: It is used to send the frame back to WebRTC native source.
3. VideoSource: It reads the camera device, produces VideoFrames, and delivers them to VideoSinks.
4. VideoProcessor: It is an interface provided by WebRTC to update videoFrames produced by videoSource .
5. MediaStream: It is an API related to WebRTC which provides support for streaming audio and video data. It consists of zero or more MediaStreamTrack objects, representing various audio or video tracks

Implement in code

Create a class to get VideoFrame to separate background

import Foundation
import WebRTC

@objc public class RTCVideoPipe: NSObject, RTCVideoCapturerDelegate {
    var virtualBackground: RTCVirtualBackground?
    var videoSource: RTCVideoSource?
    var latestTimestampNs: Int64 = 0
    var frameCount: Int = 0
    var lastProcessedTimestamp: Int64 = 0
    var fpsInterval: Int64 = 1000000000 / 15 // 15 fps to ensure VNGen can proccess
    var backgroundImage: UIImage?

    @objc public init(videoSource: RTCVideoSource) {
        self.videoSource = videoSource
        self.virtualBackground = RTCVirtualBackground()
        super.init()
    }
    
    @objc public func setBackgroundImage(image: UIImage?) {
        backgroundImage = image
    }

    @objc public func capturer(_ capturer: RTCVideoCapturer, didCapture frame: RTCVideoFrame) {
        let currentTimestamp = frame.timeStampNs

        // Calculate the time since the last processed frame
        let elapsedTimeSinceLastProcessedFrame = currentTimestamp - lastProcessedTimestamp

        if elapsedTimeSinceLastProcessedFrame < fpsInterval {
            // Skip processing the frame if it's too soon
            return
        }
        
        if backgroundImage == nil {
            self.videoSource?.emitFrame(frame)
            return
        }

        virtualBackground?.processForegroundMask(from: frame, backgroundImage: backgroundImage!) { processedFrame, error in
            if let error = error {
                // Handle error
                print("Error processing foreground mask: \(error.localizedDescription)")
            } else if let processedFrame = processedFrame {
                self.lastProcessedTimestamp = currentTimestamp

                if processedFrame.timeStampNs <= self.latestTimestampNs {
                    // Skip emitting frame if its timestamp is not newer than the latest one
                    return
                }

                self.latestTimestampNs = processedFrame.timeStampNs
                self.videoSource?.emitFrame(processedFrame)
            }
        }
    }
}

Set delegate VideoSouce to getting the VideoFrame from WebRTC

videoPipe = [[RTCVideoPipe alloc] initWithVideoSource: videoSource];
        
[videoSource setDelegate:videoPipe];

Create a class implement VisionKit to separate background

import Foundation
import AVFoundation
import Vision
import VisionKit
import OpenGLES

@available(iOS 17.0, *)
var maskRequest: VNGeneratePersonInstanceMaskRequest?

@objc public class RTCVirtualBackground: NSObject {
    
    public typealias ForegroundMaskCompletion = (RTCVideoFrame?, Error?) -> Void
    
    public override init() {
        if #available(iOS 17.0, *) {
            DispatchQueue.main.async {
                maskRequest = VNGeneratePersonInstanceMaskRequest()
            }
        }
    }
    
    public func processForegroundMask(from videoFrame: RTCVideoFrame, backgroundImage: UIImage, completion: @escaping ForegroundMaskCompletion) {
        guard let pixelBuffer = convertRTCVideoFrameToPixelBuffer(videoFrame) else {
            print("Failed to convert RTCVideoFrame to CVPixelBuffer")
            return
        }
        DispatchQueue.main.async(execute: {
            if #available(iOS 17.0, *) {
                let inputFrameImage = CIImage(cvPixelBuffer: pixelBuffer).resize()
                
                let handler = VNImageRequestHandler(ciImage: inputFrameImage!, options: [:])
                do {
                    try handler.perform([maskRequest!])
                    if let observation = maskRequest!.results?.first {
                        let allInstances = observation.allInstances
                        do {
                            let maskedImage = try observation.generateMaskedImage(ofInstances: allInstances, from: handler, croppedToInstancesExtent: false)
                            
                            self.applyForegroundMask(to: maskedImage, backgroundImage: backgroundImage) { maskedPixelBuffer, error in
                                if let maskedPixelBuffer = maskedPixelBuffer {
                                    let frameProcessed = self.convertPixelBufferToRTCVideoFrame(maskedPixelBuffer, rotation: videoFrame.rotation, timeStampNs: videoFrame.timeStampNs)
                                    completion(frameProcessed, nil)
                                } else {
                                    completion(nil, error)
                                }
                            }
                        } catch {
                            print("Error: \(error.localizedDescription)")
                            completion(nil, error)
                        }
                    }
                } catch {
                    print("Failed to perform Vision request: \(error)")
                    completion(nil, error)
                }
            }
        })   
    }
}

Draw segmented and background on canvas

func applyForegroundMask(to pixelBuffer: CVPixelBuffer, backgroundImage: UIImage, completion: @escaping (CVPixelBuffer?, Error?) -> Void) {
        DispatchQueue.global(qos: .userInitiated).async {
            let maskedUIImage = UIImage(ciImage: CIImage(cvPixelBuffer: pixelBuffer))
            
            let size = CGSize(width: CGFloat(CVPixelBufferGetWidth(pixelBuffer)), height: CGFloat(CVPixelBufferGetHeight(pixelBuffer)))
            
            let rotatedBackgroundImage = backgroundImage.rotateImage(orientation: UIImage.Orientation.up)
            
            UIGraphicsBeginImageContextWithOptions(size, false, 0.0)
            rotatedBackgroundImage.draw(in: CGRect(x: 0, y: 0, width: size.width, height: size.height))
            maskedUIImage.draw(in: CGRect(x: 0, y: 0, width: size.width, height: size.height))
            let composedImage = UIGraphicsGetImageFromCurrentImageContext()
            UIGraphicsEndImageContext()
            
            DispatchQueue.main.async {
                if let composedImage = composedImage {
                    guard let composedPixelBuffer = self.pixelBufferFromImage(image: composedImage) else {
                        completion(nil, nil)
                        return
                    }
                    
                    completion(composedPixelBuffer, nil)
                }
            }
        }
    }

Reference

• https://www.100ms.live/blog/webrtc-virtual-background-android