Priyanka Agarwal's profile - activity

Below is the code: Imgproc.matchTemplate(image, template, result, match_method); Imgproc.threshold(result, result,0.8, 1.0, Imgproc.THRESH_TOZERO);

    //Core.normalize(result, result, 0, 255, Core.NORM_MINMAX, -1, new Mat());
    // Localizing the best match with minMaxLoc
    Core.MinMaxLocResult mmr = Core.minMaxLoc(result);
    Point matchLoc;
    if (match_method == Imgproc.TM_SQDIFF || match_method == Imgproc.TM_SQDIFF_NORMED) {
        matchLoc = mmr.minLoc;
    } else {
        matchLoc = mmr.maxLoc;
    }

    //Rendering the rectangle around the template
    double tolerance = 0.80;
    Point pt2 = new Point(matchLoc.x + template.cols(), matchLoc.y + template.rows());
    Scalar color = new Scalar(0, 255, 0, 255);
    if (mmr.maxVal >= tolerance)
        Imgproc.rectangle(image, matchLoc, pt2, color);

I managed drawing the rectangle only when my image is in camera preview, but that is very hard to detect. As well as when I rotate this image nothing happens, is this the correct algorithm to detect an image. The frames per second is 1, which makes it very slow. How can I fix these issues?

Hi Steven, I am trying to match a template in camera frame, but even before I bring the template in camera frame , a rectangle is displayed.However, once the template is in camera frame,, then the rectangle is drawn only around the template. So I wanted to draw the rectangle only when template is found. Any suggestion will be of great help.

A rectangle is drawn irrespective of the template found or not. How can I know taht a valid match for the template has been found?

Thank you so much for the input. All I want is to detect a particular image (similar to ImageTargets in Vuforia SDK). So what would be the correct way to do so. Please suggest, as I am blocked at cascade generation with one positive image only.

I get it now, I have to have negatives, which can be anything other than my image. What should be the number of negatives when I have only one positive image?

thank you, but what should I do to detect an image (one particular image), which can be anywhere, on a wall, a desk, a bottle, etc. Then where do I get these negatives from? Any suggestion would be of great help.

What value should we use as -numNeg and -numPos. Why are negatives needed at all? For example, When I am trying to detect a particular car image, and not sure about the background,why is negative dataset required? Can I generate a cascade file with no negative images and one car image?

Hi Berak, I used the below commands to generate one cascade file but card detection is very hard.. 1. Generate positive samples using 1 image – \opencv_createsamples -img car.png -bg bg.txt -info annotations.lst -maxxangle 0.1 -maxyangle 0.1 -maxzangle 0.1
2. Generate vec file from the annotation.lst – \opencv_createsamples -vec car.vec -info annotations.lst -maxxangle 0.1 -maxyangle 0.1 -maxzangle 0.1
3. View the .vec file – \opencv_createsamples -vec car.vec -w 24 -h 24 4. Train the cascade,generate xml – \opencv_traincascade -data carcascade -vec car.vec -bg bg.txt -nstages 10 -minhitrate 0.999 -maxfalsealarm 0.5 -featureType LBP -numPos 100 -numNeg 100 -w 24 -h 24.

So it means that with one image, I will not be able to generate a good cascade file. How many positive and negative images are actually needed, is there some proportion which I need to maintain?

I have an image of a company's logo and now I want to detect it. This logo can be present anywhere, on top of any thing. As soon as I get this logo in my camera feed, I want to detect it and mark it.

Thanks Berak. Yes I am trying with one positive, because my goal is to detect a particular image. Is that not possible with OpenCV?

I am going through many other links and materials, but any suggestion will be of great help.

I tried generating a cascade file for detecting a car. I have 500 images in negative folder and one positive image. I used the below commands: 1. opencv_createsamples -img car.png -num 550 -w 48 -h 24 –vec cars.vec 2. 2. opencv_traincascade -data data -vec cars.vec -bg bg.txt -numStages 10 -minhitrate 0.999 -maxfalsealarm 0.1 -featureType LBP -numPos 1 -numNeg 500 -w 48 -h 24 3. I replaced the cascade file in face-detection sample app with this new generated cascade file.

But many false positives gets detected too along with the car. Probably the cascade file which I generated is not correct enough to detect only the car which I supplied as a positive image.

Thanks for the response Berak. Can you help me get the steps to generate a correct cascade file?

I have installed opencv-3.0.0 using the installer from "http://downloads.sourceforge.net/opencvlibrary/opencv-3.0.0.exe". I am running the utilities opencv_createsamples and opencv_traincascade directly after running the installer. Is this the correct way to run the utilities?