Detecting colorcheckers using neural network

In this tutorial you will learn how to use the neural network to boost up the accuracy of the chart detection algorithm.

Building

When building OpenCV, run the following command to build all the contrib module:

cmake -D OPENCV_EXTRA_MODULES_PATH=<opencv_contrib>/modules/

Or only build the mcc module:

cmake -D OPENCV_EXTRA_MODULES_PATH=<opencv_contrib>/modules/mcc

Or make sure you check the mcc module in the GUI version of CMake: cmake-gui.

Source Code of the sample

You can run the sample code by doing

<path_of_your_opencv_build_directory>/bin/example_mcc_chart_detection_with_network -t=<type_of_chart> -m=<path_to_neural_network> -pb=<path_to_models_pbtxt> -v=<optional_path_to_video_if_not_provided_webcam_will_be_used.mp4> --ci=<optional_camera_id_needed_only_if_video_not_provided> --nc=<optional_maximum_number_of_charts_in_image> --use_gpu <optional_should_gpu_be_used>
 
``'
 
* -t=#  is the chart type where 0 (Standard), 1 (DigitalSG), 2 (Vinyl)
* --ci=#  is the camera ID where 0 (default is the main camera), 1 (secondary camera) etc
* --nc=#  By default its values is 1 which means only the best chart will be detected
 
Example:

Simple run on CPU (GPU wont be used) /home/opencv/build/bin/example_mcc_chart_detection_with_network -t=0 -m=/home/model.pb –pb=/home/model.pbtxt -v=mcc24.mp4

To run on GPU /home/opencv/build/bin/example_mcc_chart_detection_with_network -t=0 -m=/home/model.pb –pb=/home/model.pbtxt -v=mcc24.mp4 –use_gpu

To run on GPU and detect the best 5 charts (Detections can be less than 5 but not more than 5) /home/opencv/build/bin/example_mcc_chart_detection_with_network -t=0 -m=/home/model.pb –pb=/home/model.pbtxt -v=mcc24.mp4 –use_gpu –nc=5 ```

#include <opencv2/core.hpp>
 
#include <opencv2/highgui.hpp>
#include <opencv2/mcc.hpp>
#include <opencv2/dnn.hpp>
#include <iostream>
 
using namespace std;
using namespace cv;
using namespace mcc;
 
const char *about = "Basic chart detection using neural network";
const char *keys = {
    "{ help h usage ? |    | show this message }"
    "{t       | 0   | chartType: 0-Standard, 1-DigitalSG, 2-Vinyl, default:0}"
    "{m       |     | File path of model, if you don't have the model you can \
                      find the link in the documentation}"
    "{pb      |     | File path of pbtxt file, available along with with the model \
                      file }"
    "{v       |     | Input from video file, if ommited, input comes from camera }"
    "{ci      | 0   | Camera id if input doesnt come from video (-v) }"
    "{nc      | 1   | Maximum number of charts in the image }"
    "{use_gpu |     | Add this flag if you want to use gpu}"};
 
int main(int argc, char *argv[])
{
    // ----------------------------------------------------------
    // Scroll down a bit (~50 lines) to find actual relevant code
    // ----------------------------------------------------------
 
    CommandLineParser parser(argc, argv, keys);
    parser.about(about);
 
    if (parser.has("help"))
    {
        parser.printMessage();
        return -1;
    }
 
    int t = parser.get<int>("t");
 
    CV_Assert(0 <= t && t <= 2);
    TYPECHART chartType = TYPECHART(t);
 
    string model_path = parser.get<string>("m");
    string pbtxt_path = parser.get<string>("pb");
 
    int camId = parser.get<int>("ci");
    int nc = parser.get<int>("nc");
 
    String video;
 
    if (parser.has("v"))
        video = parser.get<String>("v");
 
    bool use_gpu = parser.has("use_gpu");
 
    if (!parser.check())
    {
        parser.printErrors();
        return 0;
    }
 
    VideoCapture inputVideo;
    int waitTime;
    if (!video.empty())
    {
        inputVideo.open(video);
        waitTime = 10;
    }
    else
    {
        inputVideo.open(camId);
        waitTime = 10;
    }
 
    //--------------------------------------------------------------------------
    //-------------------------Actual Relevant Code-----------------------------
    //--------------------------------------------------------------------------
 
    //load the network
 
    cv::dnn::Net net = cv::dnn::readNetFromTensorflow(model_path, pbtxt_path);
 
    if (use_gpu)
    {
        net.setPreferableBackend(dnn::DNN_BACKEND_CUDA);
        net.setPreferableTarget(dnn::DNN_TARGET_CUDA);
    }
 
    Ptr<CCheckerDetector> detector = CCheckerDetector::create();
    if (!detector->setNet(net))
    {
        cout << "Loading Model failed: Aborting" << endl;
        return 0;
    }
 
    while (inputVideo.grab())
    {
        Mat image, imageCopy;
        inputVideo.retrieve(image);
 
        imageCopy = image.clone();
 
        // Marker type to detect
        if (!detector->process(image, chartType, nc, true))
        {
            printf("ChartColor not detected \n");
        }
        else
        {
 
            // get checker
            std::vector<Ptr<mcc::CChecker>> checkers = detector->getListColorChecker();
            for (Ptr<mcc::CChecker> checker : checkers)
            {
                // current checker
 
                Ptr<CCheckerDraw> cdraw = CCheckerDraw::create(checker);
                cdraw->draw(image);
            }
        }
 
        imshow("image result | q or esc to quit", image);
        imshow("original", imageCopy);
        char key = (char)waitKey(waitTime);
        if (key == 27)
            break;
    }
 
    return 0;
}

Explanation

1. Set header and namespaces

   #include <opencv2/mcc.hpp>
   using namespace std;
   using namespace cv;
   using namespace mcc;

   If you want you can set the namespace like the code above.

2. Create the detector object

   Ptr<CCheckerDetector> detector = CCheckerDetector::create();

   This is just to create the object.

3. Load the model

   cv::dnn::Net net = cv::dnn::readNetFromTensorflow(model_path, pbtxt_path);

   Load the model, here the model supplied with model was trained in tensorflow so we are loading it in tensorflow, but if you have some other model trained in some other framework you can use that also.

4. (Optional) Set the dnn backend to CUDA

   net.setPreferableBackend(dnn::DNN_BACKEND_CUDA);
   net.setPreferableTarget(dnn::DNN_TARGET_CUDA);

   Models run much faster on CUDA, so use CUDA if possible.

5. Run the detector

   detector->process(image, chartType, max_number_charts_in_image, true);

   If the detector successfully detects atleast one chart, it return true otherwise it returns false. In the above given code we print a failure message if no chart were detected. Otherwise if it were successful, the list of colorcharts is stored inside the detector itself, we will see in the next step on how to extract it. The fourth parameter is for deciding whether to use the net or not.

6. Get List of ColorCheckers

   std::vector<cv::Ptr<mcc::CChecker>> checkers;
   detector->getListColorChecker(checkers);

   All the colorcheckers that were detected are now stored in the 'checkers' vector.

7. Draw the colorcheckers back to the image

   for(Ptr<mcc::CChecker> checker : checkers)
   {
       // current checker
       Ptr<CCheckerDraw> cdraw = CCheckerDraw::create(checker);
       cdraw->draw(image);
   }

   Loop through all the checkers one by one and then draw them.