ML-KIT Face Detector无法检测人脸问题求助

1. 修正图像旋转角度 #

CameraX的ImageProxy自带传感器旋转信息，硬编码90度旋转会导致部分设备上人脸方向偏移/颠倒，ML-Kit无法识别。应使用imageProxy.imageInfo.rotationDegrees获取正确旋转值：

override fun onCaptureSuccess(imageProxy: ImageProxy) {
    val bitmap = imageProxy.image?.let { BitmapUtils().convertProxyImageInBitMap(it) }
    bitmap?.let {
        val rotationDegrees = imageProxy.imageInfo.rotationDegrees
        val photo = BitmapUtils().rotateBitmap(it, rotationDegrees.toFloat())
        detectFaces(photo, returnIntent)
    }
    // 必须关闭ImageProxy，避免CameraX资源泄漏
    imageProxy.close()
}

2. 跳过Bitmap转换，直接使用MediaImage #

Bitmap转换过程可能导致图像格式失真或信息丢失，推荐直接用ImageProxy创建InputImage，效率更高且检测更准确：

override fun onCaptureSuccess(imageProxy: ImageProxy) {
    val inputImage = InputImage.fromMediaImage(imageProxy.image!!, imageProxy.imageInfo.rotationDegrees)
    detectFaces(inputImage, returnIntent)
    imageProxy.close()
}

// 调整detectFaces函数参数
private fun detectFaces(inputImage: InputImage, intent: Intent){
    if (::faceDetector.isInitialized) {
        faceDetector.process(inputImage)
            .addOnSuccessListener { faces ->
                if (faces.size >= 1) {
                    // 处理人脸检测成功逻辑
                }
            }
            .addOnFailureListener {
                it.printStackTrace()
            }
    }
}

3. 检查Bitmap转换工具类 #

如果必须使用Bitmap，确保convertProxyImageInBitMap正确处理CameraX默认输出的YUV_420_888格式，示例实现：

fun convertProxyImageInBitMap(image: Image): Bitmap {
    val yBuffer = image.planes[0].buffer
    val uBuffer = image.planes[1].buffer
    val vBuffer = image.planes[2].buffer

    val ySize = yBuffer.remaining()
    val uSize = uBuffer.remaining()
    val vSize = vBuffer.remaining()

    val nv21 = ByteArray(ySize + uSize + vSize)
    yBuffer.get(nv21, 0, ySize)
    vBuffer.get(nv21, ySize, vSize)
    uBuffer.get(nv21, ySize + vSize, uSize)

    val yuvImage = YuvImage(nv21, ImageFormat.NV21, image.width, image.height, null)
    val outputStream = ByteArrayOutputStream()
    yuvImage.compressToJpeg(Rect(0, 0, image.width, image.height), 100, outputStream)
    val imageBytes = outputStream.toByteArray()
    return BitmapFactory.decodeByteArray(imageBytes, 0, imageBytes.size)
}

4. 优化FaceDetector配置 #

调整检测器参数，确保不会过滤掉小尺寸人脸：

faceDetectorOptions = FaceDetectorOptions.Builder()
    .setPerformanceMode(FaceDetectorOptions.PERFORMANCE_MODE_ACCURATE)
    .setContourMode(FaceDetectorOptions.CONTOUR_MODE_ALL)
    .setLandmarkMode(FaceDetectorOptions.LANDMARK_MODE_ALL)
    .setClassificationMode(FaceDetectorOptions.CLASSIFICATION_MODE_ALL)
    .setMinFaceSize(0.15f) // 人脸占图像比例至少15%，可根据需求调整
    .build()

内容的提问来源于stack exchange，提问作者Zannardy A. Oliveira