如何将face-api.js图片人脸识别改造为webcam实时识别且不使用Node.js

1. 替换HTML代码 #

移除原有的文件上传input组件，替换为视频播放标签，完整HTML代码如下：

<html lang="en">
<head>
  <meta charset="UTF-8">
  <meta name="viewport" content="width=device-width, initial-scale=1.0">
  <meta http-equiv="X-UA-Compatible" content="ie=edge">

  <script defer src="face-api.min.js"></script>
  <script defer src="script.js"></script>

  <title>Real-time Face Recognition</title>
  <style>
    body {
      margin: 0;
      padding: 0;
      width: 100vw;
      height: 100vh;
      display: flex;
      justify-content: center;
      align-items: center;
      flex-direction: column
    }

    canvas {
      position: absolute;
      top: 0;
      left: 0;
    }
  </style>
</head>
<body>
  <video id="video" autoplay muted playsinline></video>
</body>
</html>

2. 替换script.js代码 #

核心逻辑替换为摄像头调用、逐帧人脸检测，完整代码如下：

const video = document.getElementById('video')

Promise.all([
  faceapi.nets.faceRecognitionNet.loadFromUri('/models'),
  faceapi.nets.faceLandmark68Net.loadFromUri('/models'),
  faceapi.nets.ssdMobilenetv1.loadFromUri('/models')
]).then(start)

async function start() {
  const container = document.createElement('div')
  container.style.position = 'relative'
  document.body.append(container)
  document.body.append('Models Loaded, Please Allow Camera Permission')
  
  // 加载已标记的人脸特征
  const labeledFaceDescriptors = await loadLabeledImages()
  const faceMatcher = new faceapi.FaceMatcher(labeledFaceDescriptors, 0.6)
  
  // 加载摄像头流
  const stream = await navigator.mediaDevices.getUserMedia({ video: {} })
  video.srcObject = stream
  
  // 初始化画布
  const canvas = faceapi.createCanvasFromMedia(video)
  container.append(video)
  container.append(canvas)

  video.addEventListener('loadedmetadata', () => {
    const displaySize = { width: video.videoWidth, height: video.videoHeight }
    faceapi.matchDimensions(canvas, displaySize)

    // 每100ms执行一次人脸检测
    setInterval(async () => {
      const detections = await faceapi.detectAllFaces(video)
        .withFaceLandmarks()
        .withFaceDescriptors()
      
      const resizedDetections = faceapi.resizeResults(detections, displaySize)
      // 清空上一帧的绘制内容
      canvas.getContext('2d').clearRect(0, 0, canvas.width, canvas.height)

      const results = resizedDetections.map(d => faceMatcher.findBestMatch(d.descriptor))
      results.forEach((result, i) => {
        const box = resizedDetections[i].detection.box
        const drawBox = new faceapi.draw.DrawBox(box, { label: result.toString() })
        drawBox.draw(canvas)
      })
    }, 100)
  })
}

function loadLabeledImages() {
  const labels = ['Black Widow', 'Captain America', 'Captain Marvel', 'Hawkeye', 'Jim Rhodes', 'Thor', 'Tony Stark']
  return Promise.all(
    labels.map(async label => {
      const descriptions = []
      for (let i = 1; i <= 2; i++) {
        const img = await faceapi.fetchImage(`https://raw.githubusercontent.com/WebDevSimplified/Face-Recognition-JavaScript/master/labeled_images/${label}/${i}.jpg`)
        const detections = await faceapi.detectSingleFace(img).withFaceLandmarks().withFaceDescriptor()
        descriptions.push(detections.descriptor)
      }
      return new faceapi.LabeledFaceDescriptors(label, descriptions)
    })
  )
}

注意事项 #

• 项目根目录需要保留原有的models文件夹，存放face-api所需的模型文件
• 必须在localhost或HTTPS环境下运行，VSCode的LiveServer环境符合要求，否则浏览器会禁止调用摄像头权限
• 如果出现检测卡顿，可以将代码中setInterval的间隔时间从100调大，降低检测频率

内容的提问来源于stack exchange，提问作者Ali Firoozi