FRCVision C++ MultiCamera Example with Debug Video Output

[phurley67]

We are working with the C++ multiCameraServer with a GRIP pipeline. Within the loop we have a simple wrapper around the generated GRIP pipeline (where grep::InfinteRecharge is an unmodified C++ pipeline generated by GRIP):

class Pipeline : public grip::InfiniteRecharge, cs::CvSource  {
public:
   Pipeline(const wpi::Twine& name, const cs::VideoMode& mode) : cs::CvSource(name, mode) { }

   void Process(cv::Mat& mat) override {
      grip::InfiniteRecharge::Process(mat);
      PutFrame(*GetCvErodeOutput());
   }
};

And then in the main there is some code like:

    SwitchedCameraConfig debugOutput = { "debug", "debug" };
    Pipeline* pipeline = new Pipeline("debug", cameras[0].GetVideoMode());
    auto debugCamera = StartSwitchedCamera(debugOutput);

    if (cameras.size() >= 1) {
        std::thread([&] {
           frc::VisionRunner<Pipeline> runner(cameras[0], pipeline, [&](Pipeline &pipeline) {
                  *snip some contour bounding box code, that writes to network tables

Just trying to add a second network stream that would allow us to watch a masked stream (or any other cv::Mat we wanted to see.

Am I going the wrong direction? (just trying to keep things working with the framework, but not quite sure where to add the output).

Thanks for your time

[Peter_Johnson]

To create a new stream that you can send OpenCV images to, just call CameraServer::GetInstance().PutVideo(). This creates a MjpegServer and a CvSource and hooks one to the other, and returns the CvSource. I generally recommend using composition over inheritance (e.g. having a member of your Pipeline class be cs::CvSource instead of deriving from it), but if you want to keep your current structure, just change the cs::CvSource() portion of the construction to be cs::CvSource(CameraServer::GetInstance().PutVideo(...)).

You can also create a MjpegServer yourself and hook up a CvSource manually to it using SetSource(), but I recommend using the CameraServer convenience functions as they also publish information about the created stream to NetworkTables so they show up in dashboard camera lists.

[phurley67]

Thank you. Moved the camera initialization out of the pipeline completely. Switch it on/off now and have two extra streams when enabled for debugging.

Works great.