Opencv 代码 - 有人可以帮我理解代码在做啥吗

Posted 2023-02-21

【中文标题】Opencv 代码 - 有人可以帮我理解代码在做啥吗

【英文标题】：Open CV Code - Can someone help me understand what the code is doingOpencv 代码 - 有人可以帮我理解代码在做什么吗

【发布时间】：2015-06-28 12:47:31

【问题描述】：

我正在研究 OpenCV 颜色 blob，并且正在查看代码。我需要一些帮助来了解代码在做什么，以便我可以进一步处理它并帮助进一步集成它。是否有人可以帮助我理解/评论代码，以便更容易进行交互。

public boolean onTouch(View v, MotionEvent event) 
        int cols = mRgba.cols(); //get resolution of display
        int rows = mRgba.rows(); // get resolution of display

        int xOffset = (mOpenCvCameraView.getWidth() - cols) / 2; //get resolution of display
        int yOffset = (mOpenCvCameraView.getHeight() - rows) / 2; // get resolution of display

        int x = (int)event.getX() - xOffset; // get resolution of display
        int y = (int)event.getY() - yOffset; //get resolution of display 

        Log.i(TAG, "Touch image coordinates: (" + x + ", " + y + ")");

        if ((x < 0) || (y < 0) || (x > cols) || (y > rows)) return false;

        Rect touchedRect = new Rect();

        touchedRect.x = (x>4) ? x-4 : 0;
        touchedRect.y = (y>4) ? y-4 : 0;

        touchedRect.width = (x+4 < cols) ? x + 4 - touchedRect.x : cols - touchedRect.x;
        touchedRect.height = (y+4 < rows) ? y + 4 - touchedRect.y : rows - touchedRect.y;

        Mat touchedRegionRgba = mRgba.submat(touchedRect);

        Mat touchedRegionHsv = new Mat();
        Imgproc.cvtColor(touchedRegionRgba, touchedRegionHsv, Imgproc.COLOR_RGB2HSV_FULL);

        // Calculate average color of touched region
        mBlobColorHsv = Core.sumElems(touchedRegionHsv);
        int pointCount = touchedRect.width*touchedRect.height;
        for (int i = 0; i < mBlobColorHsv.val.length; i++)
            mBlobColorHsv.val[i] /= pointCount;

        //converts scalar to hsv to RGB
        mBlobColorRgba = converScalarHsv2Rgba(mBlobColorHsv);

        Log.i(TAG, "Touched rgba color: (" + mBlobColorRgba.val[0] + ", " + mBlobColorRgba.val[1] +
                ", " + mBlobColorRgba.val[2] + ", " + mBlobColorRgba.val[3] + ")");

        mDetector.setHsvColor(mBlobColorHsv);

        Imgproc.resize(mDetector.getSpectrum(), mSpectrum, SPECTRUM_SIZE);

        mIsColorSelected = true;

        touchedRegionRgba.release();
        touchedRegionHsv.release();

        return false; // don't need subsequent touch events
    

    public Mat onCameraFrame(CvCameraViewFrame inputFrame) 
        mRgba = inputFrame.rgba();

        if (mIsColorSelected) 
            mDetector.process(mRgba);
            List<MatOfPoint> contours = mDetector.getContours();
            Log.e(TAG, "Contours count: " + contours.size());
            Imgproc.drawContours(mRgba, contours, -1, CONTOUR_COLOR);

            Mat colorLabel = mRgba.submat(4, 68, 4, 68);
            colorLabel.setTo(mBlobColorRgba);

            Mat spectrumLabel = mRgba.submat(4, 4 + mSpectrum.rows(), 70, 70 + mSpectrum.cols());
            mSpectrum.copyTo(spectrumLabel);
        

        return mRgba;
    
//final conversion
    private Scalar converScalarHsv2Rgba(Scalar hsvColor) 
        Mat pointMatRgba = new Mat();
        Mat pointMatHsv = new Mat(1, 1, CvType.CV_8UC3, hsvColor);
        Imgproc.cvtColor(pointMatHsv, pointMatRgba, Imgproc.COLOR_HSV2RGB_FULL, 4);

        return new Scalar(pointMatRgba.get(0, 0));
    

【参考方案1】：

我已尽我所能浏览并评论了代码，尽管我认为它是相当自我记录的。 通常，此代码从用户触摸绘制的坐标中获取图像。然后它转换颜色空间 并根据频谱大小调整其大小

    // When a motion event happens (someone touches the device)
public boolean onTouch(View v, MotionEvent event) 
    int cols = mRgba.cols(); //get resolution of display
    int rows = mRgba.rows(); // get resolution of display

    int xOffset = (mOpenCvCameraView.getWidth() - cols) / 2; //get resolution of display
    int yOffset = (mOpenCvCameraView.getHeight() - rows) / 2; // get resolution of display

    int x = (int)event.getX() - xOffset; // get resolution of display
    int y = (int)event.getY() - yOffset; //get resolution of display 

        //The place where the screen was touched
    Log.i(TAG, "Touch image coordinates: (" + x + ", " + y + ")");
    // ensure it is within the screen.
    if ((x < 0) || (y < 0) || (x > cols) || (y > rows)) return false;

    Rect touchedRect = new Rect();

    //Ensure it is a multiple of 4
    touchedRect.x = (x>4) ? x-4 : 0;
    touchedRect.y = (y>4) ? y-4 : 0;

    // If  x+4 < cols then ?"" else :""
    touchedRect.width = (x+4 < cols) ? x + 4 - touchedRect.x : cols - touchedRect.x;
    touchedRect.height = (y+4 < rows) ? y + 4 - touchedRect.y : rows - touchedRect.y;

    create a touched regionmat from the image created from the touches
    Mat touchedRegionRgba = mRgba.submat(touchedRect);

    //Convert the new mat to HSV colour space
    Mat touchedRegionHsv = new Mat();
    Imgproc.cvtColor(touchedRegionRgba, touchedRegionHsv, Imgproc.COLOR_RGB2HSV_FULL);

    // Calculate average color of touched region
    mBlobColorHsv = Core.sumElems(touchedRegionHsv);
    int pointCount = touchedRect.width*touchedRect.height;
    for (int i = 0; i < mBlobColorHsv.val.length; i++)
        mBlobColorHsv.val[i] /= pointCount;

    //converts scalar to hsv to RGB
    mBlobColorRgba = converScalarHsv2Rgba(mBlobColorHsv);

    Log.i(TAG, "Touched rgba color: (" + mBlobColorRgba.val[0] + ", " + mBlobColorRgba.val[1] +
            ", " + mBlobColorRgba.val[2] + ", " + mBlobColorRgba.val[3] + ")");

    mDetector.setHsvColor(mBlobColorHsv);

    // Resize the image to specture size
    Imgproc.resize(mDetector.getSpectrum(), mSpectrum, SPECTRUM_SIZE);

    mIsColorSelected = true;

    // Release all mats
    touchedRegionRgba.release();
    touchedRegionHsv.release();

    return false; // don't need subsequent touch events


public Mat onCameraFrame(CvCameraViewFrame inputFrame) 
    mRgba = inputFrame.rgba();

    if (mIsColorSelected) 
        mDetector.process(mRgba);
        List<MatOfPoint> contours = mDetector.getContours();
        Log.e(TAG, "Contours count: " + contours.size());
        Imgproc.drawContours(mRgba, contours, -1, CONTOUR_COLOR);

        Mat colorLabel = mRgba.submat(4, 68, 4, 68);
        colorLabel.setTo(mBlobColorRgba);

        Mat spectrumLabel = mRgba.submat(4, 4 + mSpectrum.rows(), 70, 70 + mSpectrum.cols());
        mSpectrum.copyTo(spectrumLabel);
    

    return mRgba;

//final conversion
    private Scalar converScalarHsv2Rgba(Scalar hsvColor) 
        Mat pointMatRgba = new Mat();
        Mat pointMatHsv = new Mat(1, 1, CvType.CV_8UC3, hsvColor);
        Imgproc.cvtColor(pointMatHsv, pointMatRgba, Imgproc.COLOR_HSV2RGB_FULL, 4);

        return new Scalar(pointMatRgba.get(0, 0));
    

【参考方案2】：

对于任何对“mDetector.process(mRgba)”方法的作用感兴趣的人，以下可能会有用。

作为另一个答案的回顾，当用户触摸一个区域时，程序会检测他们所触摸区域的颜色。

每次新的相机帧到达时，程序都会检查是否选择了颜色，如果是，则调用方法“mDetector.process(mRgba)”来处理帧。

此方法在下面注释，但高级步骤是：

缩小图片尺寸 过滤掉所有不是我们正在寻找的颜色的东西 增强图像的其余部分，以便更轻松地检测边缘或轮廓 找到剩余颜色“斑点”的轮廓（轮廓或边缘） 过滤掉任何太小而无法感兴趣的内容 返回剩余斑点的列表，或更准确地说，返回作为剩余斑点轮廓的轮廓列表

注释代码（android）：

 public void process(Mat rgbaImage) 

        //Pyramid Down - this downsizes the image and looses some resolution
        //See: http://docs.opencv.org/2.4/doc/tutorials/imgproc/pyramids/pyramids.html
        Imgproc.pyrDown(rgbaImage, mPyrDownMat);
        Imgproc.pyrDown(mPyrDownMat, mPyrDownMat);

        //Convert color scheme to HSV - this means that a color can be
        //identified with a single value, the hue, instead of three values
        Imgproc.cvtColor(mPyrDownMat, mHsvMat, Imgproc.COLOR_RGB2HSV_FULL);

        //This creates a new image with only the color values that are wihtin
        //the lower and upper thresholds set in mLowerBound and mUpperBound. These
        //values were calculated when the method 'setHsvColor' was called with the
        //color of the object that the user touched on the screen.
        //So you effectively get an image with just the red or just the blue or whatever
        //the color of the blob that the user selected was. Note that if there are multiple
        //blobs or objects with this color you will get them all. You can see this quite easily
        //with a simple test of the app with a couple of similar colored objects.
        Core.inRange(mHsvMat, mLowerBound, mUpperBound, mMask);

        //dilate effectively emphasises the brighter colors, so making them bigger within the image
        //In this case it should be the chosen color which is emphasised against the
        //darker (black) background.
        //See:http://docs.opencv.org/2.4/doc/tutorials/imgproc/erosion_dilatation/erosion_dilatation.html
        Imgproc.dilate(mMask, mDilatedMask, new Mat());

        List<MatOfPoint> contours = new ArrayList<MatOfPoint>();

        //Finds the contours which in this case means the edge of the color blobs
        Imgproc.findContours(mDilatedMask, contours, mHierarchy, Imgproc.RETR_EXTERNAL, Imgproc.CHAIN_APPROX_SIMPLE);

        // Find max contour area
        //This is actually refering to the area enclosed by a contour. For this to work it is important
        //that the contour be closed, so if this is not the case some objects may be missed here.
        double maxArea = 0;
        Iterator<MatOfPoint> each = contours.iterator();
        while (each.hasNext()) 
            MatOfPoint wrapper = each.next();
            double area = Imgproc.contourArea(wrapper);
            if (area > maxArea)
                maxArea = area;
        

        // Filter contours by area and resize to fit the original image size
        //Here we are simply discrading any contours that are below the min size that was
        //set in the method 'setMinContourArea' or the default if it was not set. In other
        //words discrading any small object detected.
        mContours.clear();
        each = contours.iterator();
        while (each.hasNext()) 
            MatOfPoint contour = each.next();
            if (Imgproc.contourArea(contour) > mMinContourArea*maxArea) 
                Core.multiply(contour, new Scalar(4,4), contour);
                mContours.add(contour);
            
        
    

    //Now we return the list of contours - each contour is a closed area that is
    //colored in whatever color the user selected when they touched the object.
    //This color, as a reminder, was set by a call to 'setHsvColor'.
    public List<MatOfPoint> getContours() 
        return mContours;