11
13

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 3 years have passed since last update.

【Swift】Vision.frameworkでカメラ画像の顔認識を行う【iOS】

Last updated at Posted at 2019-12-21

iOS11より、iOS標準フレームワーク Vision.framework を使うと、顔認識ができるらしいので今更ながら使ってみました。

概要

カメラ画像から顔を検出し、顔部分に矩形を表示します。

試した環境

  • Xcode 11.3
  • iOS 13.2
  • swift 5

実行サンプル

ぱくたそフリー素材で実験

IMG_3057.jpg
ディスプレイ画質の問題のせいもありそうですが、顔にちょっと髪がかかってたりすると少し認識が悪い。

Google画像検索「顔」で実験

IMG_3059.jpg
顔が沢山あっても、アップだと良く認識します。
(画像はぼかしてます

コード説明

VNImageRequestHandler
を利用して、 pixelBuffer から、顔情報を配列取得します。

結果は
VNDetectFaceRectanglesRequest
に非同期で戻されます。
顔情報は VNFaceObservation です。

    /// 顔認識情報の配列取得 (非同期)
    private func getFaceObservations(pixelBuffer: CVPixelBuffer, completion: @escaping (([VNFaceObservation])->())) {
        let request = VNDetectFaceRectanglesRequest { (request, error) in
            guard let results = request.results as? [VNFaceObservation] else {
                completion([])
                return
            }
            completion(results)
        }

        let handler = VNImageRequestHandler(cvPixelBuffer: pixelBuffer, options: [:])
        try? handler.perform([request])
    }

pixcelBuffer は カメラから取得した sampleBufferCMSampleBufferGetImageBuffer を使って変換します。
imageView には、 sampleBuffer から取得した生成をセット。

    /// カメラからの映像取得デリゲート
    func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
        guard let pixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else {
            return
        }
        getFaceObservations(pixelBuffer: pixelBuffer) { [weak self] faceObservations in
            guard let self = self else { return }
            let image = self.getFaceRectsImage(sampleBuffer: sampleBuffer, faceObservations: faceObservations)
            DispatchQueue.main.async { [weak self] in
                self?.previewImageView.image = image
            }
        }
    }

またその際、 VNFaceObservation から正規化された画像の位置が取得できるので、
その情報をもとに、矩形を画像に書き込みます。

        let imageSize = CGSize(width: width, height: height)
        let faseRects = faceObservations.compactMap {
            getUnfoldRect(normalizedRect: $0.boundingBox, targetSize: imageSize)
        }
        faseRects.forEach{ self.drawRect($0, context: newContext) }
    /// 正規化された矩形位置を指定領域に展開
    private func getUnfoldRect(normalizedRect: CGRect, targetSize: CGSize) -> CGRect {
        return CGRect(
            x: normalizedRect.minX * targetSize.width,
            y: normalizedRect.minY * targetSize.height,
            width: normalizedRect.width * targetSize.width,
            height: normalizedRect.height * targetSize.height
        )
    }
    /// コンテキストに矩形を描画
    private func drawRect(_ rect: CGRect, context: CGContext) {
        context.setLineWidth(4.0)
        context.setStrokeColor(UIColor.green.cgColor)
        context.stroke(rect)
    }

コード全体

import UIKit
import AVFoundation
import Vision

class FaceViewController: UIViewController {
 
    @IBOutlet weak var previewImageView: UIImageView!
    
    private let avCaptureSession = AVCaptureSession()
 
    override func viewDidLoad() {
        super.viewDidLoad()
        setupCamera()
    }
 
    override func viewDidDisappear(_ animated: Bool) {
        super.viewDidDisappear(animated)
        self.avCaptureSession.stopRunning()
    }
 
    /// カメラのセットアップ
    private func setupCamera() {
        self.avCaptureSession.sessionPreset = .photo
 
        let device = AVCaptureDevice.default(for: .video)
        let input = try! AVCaptureDeviceInput(device: device!)
        self.avCaptureSession.addInput(input)
 
        let videoDataOutput = AVCaptureVideoDataOutput()
        videoDataOutput.videoSettings = [kCVPixelBufferPixelFormatTypeKey as String : Int(kCVPixelFormatType_32BGRA)]
        videoDataOutput.alwaysDiscardsLateVideoFrames = true
        videoDataOutput.setSampleBufferDelegate(self, queue: .global())
        
        self.avCaptureSession.addOutput(videoDataOutput)
        self.avCaptureSession.startRunning()
    }

    /// コンテキストに矩形を描画
    private func drawRect(_ rect: CGRect, context: CGContext) {
        context.setLineWidth(4.0)
        context.setStrokeColor(UIColor.green.cgColor)
        context.stroke(rect)
    }
    
    /// 顔認識情報の配列取得 (非同期)
    private func getFaceObservations(pixelBuffer: CVPixelBuffer, completion: @escaping (([VNFaceObservation])->())) {
        let request = VNDetectFaceRectanglesRequest { (request, error) in
            guard let results = request.results as? [VNFaceObservation] else {
                completion([])
                return
            }
            completion(results)
        }
        
        let handler = VNImageRequestHandler(cvPixelBuffer: pixelBuffer, options: [:])
        try? handler.perform([request])
    }
    
    /// 正規化された矩形位置を指定領域に展開
    private func getUnfoldRect(normalizedRect: CGRect, targetSize: CGSize) -> CGRect {
        return CGRect(
            x: normalizedRect.minX * targetSize.width,
            y: normalizedRect.minY * targetSize.height,
            width: normalizedRect.width * targetSize.width,
            height: normalizedRect.height * targetSize.height
        )
    }
    
    /// 顔検出位置に矩形を描画した image を取得
    private func getFaceRectsImage(sampleBuffer :CMSampleBuffer, faceObservations: [VNFaceObservation]) -> UIImage? {
        
        guard let imageBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else {
            return nil
        }

        CVPixelBufferLockBaseAddress(imageBuffer, CVPixelBufferLockFlags(rawValue: 0))

        guard let pixelBufferBaseAddres = CVPixelBufferGetBaseAddressOfPlane(imageBuffer, 0) else {
            CVPixelBufferUnlockBaseAddress(imageBuffer, CVPixelBufferLockFlags(rawValue: 0))
            return nil
        }
            
        let width = CVPixelBufferGetWidth(imageBuffer)
        let height = CVPixelBufferGetHeight(imageBuffer)
        let bitmapInfo = CGBitmapInfo(rawValue:
            (CGBitmapInfo.byteOrder32Little.rawValue | CGImageAlphaInfo.premultipliedFirst.rawValue)
        )

        guard let newContext = CGContext(
            data: pixelBufferBaseAddres,
            width: width,
            height: height,
            bitsPerComponent: 8,
            bytesPerRow: CVPixelBufferGetBytesPerRow(imageBuffer),
            space: CGColorSpaceCreateDeviceRGB(),
            bitmapInfo: bitmapInfo.rawValue
            ) else
        {
            CVPixelBufferUnlockBaseAddress(imageBuffer, CVPixelBufferLockFlags(rawValue: 0))
            return nil
        }

        let imageSize = CGSize(width: width, height: height)
        let faseRects = faceObservations.compactMap {
            getUnfoldRect(normalizedRect: $0.boundingBox, targetSize: imageSize)
        }
        faseRects.forEach{ self.drawRect($0, context: newContext) }
        
        CVPixelBufferUnlockBaseAddress(imageBuffer, CVPixelBufferLockFlags(rawValue: 0))

        guard let imageRef = newContext.makeImage() else {
            return nil
        }
        let image = UIImage(cgImage: imageRef, scale: 1.0, orientation: UIImage.Orientation.right)

        return image
    }
}


extension FaceViewController : AVCaptureVideoDataOutputSampleBufferDelegate{
    
    /// カメラからの映像取得デリゲート
    func captureOutput(_ output: AVCaptureOutput, didOutput sampleBuffer: CMSampleBuffer, from connection: AVCaptureConnection) {
        guard let pixelBuffer = CMSampleBufferGetImageBuffer(sampleBuffer) else {
            return
        }
        getFaceObservations(pixelBuffer: pixelBuffer) { [weak self] faceObservations in
            guard let self = self else { return }
            let image = self.getFaceRectsImage(sampleBuffer: sampleBuffer, faceObservations: faceObservations)
            DispatchQueue.main.async { [weak self] in
                self?.previewImageView.image = image
            }
        }
    }
}

github

becky3/face_detection: 【Swift】Vision.frameworkでカメラ画像の顔認識を行う【iOS】
https://github.com/becky3/face_detection

参考サイト

11
13
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
11
13

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?