jrussino/cudaDetectAndComputeAsync.cpp

## cudaDetectAndComputeAsync.cpp
#include <iostream>
#include <thread>

#include <opencv2/cudafeatures2d.hpp>
#include <opencv2/cudaimgproc.hpp>
#include <opencv2/highgui.hpp>

void capture_thread(int device_num, int n_cycles)
{
  std::cout << "Starting capture thread for device " + std::to_string(device_num) + "\n" << std::flush;

  // open VideoCapture for the specified device
  cv::VideoCapture cap(device_num);
  if (!cap.isOpened())
  {
    std::cerr << "ERROR - Failed to open VideoCapture for device " + std::to_string(device_num) + "\n" << std::flush;
    return;
  }

  // each thread gets its own CUDA stream
  cv::cuda::Stream stream;


  std::cout << "Initialization complete for capture thread for device " + std::to_string(device_num) + "\n" << std::flush;
  for(int i = 1; i <= n_cycles; ++i)
  {
    // capture frame
    cv::Mat frame;
    if(!cap.read(frame))
    {
      continue;
    }

    // upload to GPU
    cv::cuda::GpuMat frame_d;
    frame_d.upload(frame, stream);

    // convert to grayscale
    cv::cuda::GpuMat frame_gray_d;
    cv::cuda::cvtColor(frame_d, frame_gray_d, CV_BGR2GRAY, 0, stream);

    // create CUDA ORB feature detector
    cv::Ptr<cv::cuda::ORB> p_orb_d = cv::cuda::ORB::create();
    p_orb_d->setBlurForDescriptor(true);

    // detect and compute features
    cv::cuda::GpuMat kp_d;
    cv::cuda::GpuMat desc_d;
    p_orb_d->detectAndComputeAsync(frame_gray_d, cv::cuda::GpuMat(), kp_d, desc_d, false, stream);

    /*
     * Do something with the features & descriptors...
     */
  }

  std::cout << "Capture thread for device " + std::to_string(device_num) + " finished with no errors\n" << std::flush;
}


int main(int argc, const char* argv[])
{
  std::cout << "Starting main program" << std::endl;

  // Start two capture threads
  std::thread capture_thread_0(capture_thread, 0, 1000);
  std::thread capture_thread_1(capture_thread, 1, 1000);

  // Wait for both threads to return
  capture_thread_0.join();
  capture_thread_1.join();

  std::cout << "Main program completed with no errors" << std::endl;
  return 0;
}
	#include <iostream>
	#include <thread>

	#include <opencv2/cudafeatures2d.hpp>
	#include <opencv2/cudaimgproc.hpp>
	#include <opencv2/highgui.hpp>

	void capture_thread(int device_num, int n_cycles)
	{
	std::cout << "Starting capture thread for device " + std::to_string(device_num) + "\n" << std::flush;

	// open VideoCapture for the specified device
	cv::VideoCapture cap(device_num);
	if (!cap.isOpened())
	{
	std::cerr << "ERROR - Failed to open VideoCapture for device " + std::to_string(device_num) + "\n" << std::flush;
	return;
	}

	// each thread gets its own CUDA stream
	cv::cuda::Stream stream;


	std::cout << "Initialization complete for capture thread for device " + std::to_string(device_num) + "\n" << std::flush;
	for(int i = 1; i <= n_cycles; ++i)
	{
	// capture frame
	cv::Mat frame;
	if(!cap.read(frame))
	{
	continue;
	}

	// upload to GPU
	cv::cuda::GpuMat frame_d;
	frame_d.upload(frame, stream);

	// convert to grayscale
	cv::cuda::GpuMat frame_gray_d;
	cv::cuda::cvtColor(frame_d, frame_gray_d, CV_BGR2GRAY, 0, stream);

	// create CUDA ORB feature detector
	cv::Ptr<cv::cuda::ORB> p_orb_d = cv::cuda::ORB::create();
	p_orb_d->setBlurForDescriptor(true);

	// detect and compute features
	cv::cuda::GpuMat kp_d;
	cv::cuda::GpuMat desc_d;
	p_orb_d->detectAndComputeAsync(frame_gray_d, cv::cuda::GpuMat(), kp_d, desc_d, false, stream);

	/*
	* Do something with the features & descriptors...
	*/
	}

	std::cout << "Capture thread for device " + std::to_string(device_num) + " finished with no errors\n" << std::flush;
	}


	int main(int argc, const char* argv[])
	{
	std::cout << "Starting main program" << std::endl;

	// Start two capture threads
	std::thread capture_thread_0(capture_thread, 0, 1000);
	std::thread capture_thread_1(capture_thread, 1, 1000);

	// Wait for both threads to return
	capture_thread_0.join();
	capture_thread_1.join();

	std::cout << "Main program completed with no errors" << std::endl;
	return 0;
	}