Android camera2.params.face 矩形放置在 canvas

Android camera2.params.face rectangle placement on canvas

我正在尝试在我的相机预览中实施人脸检测。我按照 Android 参考页面在 TextureView 中实现自定义相机预览,放置在 FrameLayout 中。此 FrameLayout 中还有一个具有清晰背景(与相机预览重叠)的 SurfaceView。每次更新相机预览时(每帧一次),我的应用程序都会将第一个 CaptureResult.STATISTICS_FACES 面部边界识别的 Rect 动态绘制到 SurfaceView 的 canvas .我的应用假定只需要识别一张脸。

我在绘制矩形时出现问题。如果我将脸保持在相机视图的中央,我会在正确的位置得到矩形,但是当我向上移动头部时,矩形会向右移动,而当我向右移动头部时,矩形会向下移动。好像 canvas 需要旋转 -90,但这对我不起作用(在下面的代码中解释)。

在我的 activity 的 onCreate():

//face recognition
rectangleView = (SurfaceView) findViewById(R.id.rectangleView);
rectangleView.setZOrderOnTop(true);
rectangleView.getHolder().setFormat(
    PixelFormat.TRANSPARENT); //remove black background from view
purplePaint = new Paint();
purplePaint.setColor(Color.rgb(175,0,255));
purplePaint.setStyle(Paint.Style.STROKE);

in TextureView.SurfaceTextureListener.onSurfaceTextureAvailable()(在封装camera.open()try{}块中:

Rect cameraBounds = cameraCharacteristics.get(
    CameraCharacteristics.SENSOR_INFO_ACTIVE_ARRAY_SIZE);
cameraWidth = cameraBounds.right;
cameraHeight = cameraBounds.bottom;

onSurfaceTextureUpdated() 内的同一个监听器中:

if (detectedFace != null && rectangleFace.height() > 0) {
        Canvas currentCanvas = rectangleView.getHolder().lockCanvas();
        if (currentCanvas != null) {
            currentCanvas.drawColor(Color.TRANSPARENT, PorterDuff.Mode.CLEAR);
            int canvasWidth = currentCanvas.getWidth();
            int canvasHeight = currentCanvas.getHeight();
            int l = rectangleFace.right;
            int t = rectangleFace.bottom;
            int r = rectangleFace.left;
            int b = rectangleFace.top;
            int left = (canvasWidth*l)/cameraWidth;
            int top  = (canvasHeight*t)/cameraHeight;
            int right = (canvasWidth*r)/cameraWidth;
            int bottom = (canvasHeight*b)/cameraHeight;

            currentCanvas.drawRect(left, top, right, bottom, purplePaint);
        }
        rectangleView.getHolder().unlockCanvasAndPost(currentCanvas);
    }

CameraCaptureSession.CameraCallback 中的方法 onCaptureCompletedCameraCaptureSession.setRepeatingRequest() looper 调用:

//May need better face recognition sdk or api
Face[] faces = result.get(CaptureResult.STATISTICS_FACES);

if (faces.length > 0)
{
    detectedFace = faces[0];
    rectangleFace = detectedFace.getBounds();
}

所有变量都在函数外实例化。

如果您不能完全理解我的问题或需要更多信息,请在此处发布类似的问题:

How can i handle the rotation issue with Preview & FaceDetection

然而,与上面的海报不同,我什至无法让 canvas 在旋转 canvas 后显示矩形,所以这不是解决方案。

我尝试使用 x=-y, y=x (left=-top, top=left) 将我的点旋转 -90 度,但它也不起作用。我觉得需要对这些点应用某种功能,但我不知道该怎么做。

关于如何解决这个问题有什么想法吗?

为了将来参考,这是我最终得到的解决方案:

设置一个名为 orientation_offset 的 class/Activity 变量:

orientation_offset = cameraCharacteristics.get(CameraCharacteristics.SENSOR_ORIENTATION);

这是相机传感器的视图(或面部检测矩形)需要旋转的角度才能正确查看。

然后,我改成onSurfaceTextureUpdated()

Canvas currentCanvas = rectangleView.getHolder().lockCanvas();
    if (currentCanvas != null) {

        currentCanvas.drawColor(Color.TRANSPARENT, PorterDuff.Mode.CLEAR);

        if (detectedFace != null && rectangleFace.height() > 0) {

            int canvasWidth = currentCanvas.getWidth();
            int canvasHeight = currentCanvas.getHeight();
            int faceWidthOffset = rectangleFace.width()/8;
            int faceHeightOffset = rectangleFace.height()/8;

            currentCanvas.save();
            currentCanvas.rotate(360 - orientation_offset, canvasWidth / 2, 
                canvasHeight / 2);

            int l = rectangleFace.right;
            int t = rectangleFace.bottom;
            int r = rectangleFace.left;
            int b = rectangleFace.top;
            int left = (canvasWidth - (canvasWidth*l)/cameraWidth)-(faceWidthOffset);
            int top  = (canvasHeight*t)/cameraHeight - (faceHeightOffset);
            int right = (canvasWidth - (canvasWidth*r)/cameraWidth) + (faceWidthOffset);
            int bottom = (canvasHeight*b)/cameraHeight + (faceHeightOffset);

            currentCanvas.drawRect(left, top, right, bottom, purplePaint);
            currentCanvas.restore();
        }
    }
    rectangleView.getHolder().unlockCanvasAndPost(currentCanvas);

我会留下这个问题,以防其他人有解决方案。