opencv cuda failed (createBackgroundSubtractorMOG2)

Robotics & Edge Computing Jetson & Embedded Systems Jetson Nano
1

i have test opencv-4.1.0 and opencv-3.4.6, but get same error, opencv build command:

cmake \
    -DCMAKE_BUILD_TYPE=Release \
    -DCMAKE_INSTALL_PREFIX=/opt/e3cloud \
    -DOPENCV_GENERATE_PKGCONFIG=YES \
    -DBUILD_PNG=ON \
    -DBUILD_TIFF=ON \
    -DBUILD_TBB=ON \
    -DBUILD_JPEG=ON \
    -DBUILD_JASPER=ON \
    -DBUILD_ZLIB=ON \
    -DBUILD_EXAMPLES=ON \
    -DBUILD_opencv_java=OFF \
    -DBUILD_opencv_nonfree=ON \
    -DBUILD_opencv_python=OFF \
    -DENABLE_PRECOMPILED_HEADERS=OFF \
    -DWITH_OPENCL=ON \
    -DWITH_OPENMP=ON \
    -DWITH_FFMPEG=ON \
    -DWITH_GSTREAMER=OFF \
    -DWITH_GSTREAMER_0_10=OFF \
    -DBUILD_PROTOBUF=OFF \
    -DWITH_PROTOBUF=ON \
    -DWITH_GTK=ON \
    -DWITH_VTK=OFF \
    -DWITH_TBB=ON \
    -DWITH_1394=OFF \
    -DWITH_OPENEXR=OFF \
    -DWITH_CUDA=ON \
    -DCUDA_ARCH_BIN=6.2 \
    -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda \
    -DCPU_BASELINE=FP16 \
    -DCUDA_ARCH_PTX="" \
    -DINSTALL_C_EXAMPLES=ON \
    -DINSTALL_TESTS=ON \
    -DENABLE_FAST_MATH=ON \
    -DCUDA_FAST_MATH=ON \
    -DENABLE_NEON=ON \
    -DWITH_CUBLAS=ON \
    -DWITH_CUFFT=ON \
    -DWITH_CUEIGEN=ON \
    -DWITH_opencv_dnn=ON \
    -DOPENCV_DNN_OPENCL=ON \
    -DOPENCV_ENABLE_NONFREE=ON \
    -DOPENCV_EXTRA_MODULES_PATH=/home/nv/compile/opencv-4.1.0/opencv_contrib-4.1.0/modules \
    ..

test code:

#include <opencv2/opencv.hpp>
#include <thread>
#include <iostream>
#include <mutex>
#include <opencv2/cudacodec.hpp>
#include <opencv2/cudabgsegm.hpp>
#include <opencv2/cudaobjdetect.hpp>

using namespace std;
std::mutex m;
void threadfunc(const char *filename)
{	
	cv::VideoCapture cap(filename);
	cv::Mat mat;
	cv::Ptr<cv::cuda::BackgroundSubtractorMOG2> mog = cv::cuda::createBackgroundSubtractorMOG2(0.0002, 16,false);
	cv::cuda::GpuMat gpuMat;
	cv::cuda::GpuMat gpuMsk;
	cv::Mat matMsk;

	while (true)
	{ 
		m.lock();
		cap>>mat;	
		gpuMat.upload(mat);
		mog->apply( gpuMat, gpuMsk );
		gpuMat.download( matMsk );
		cout << "Intial ThreadID : " << std::this_thread::get_id() << ":" << endl;
		m.unlock();
	} 
};

error output:

nv@nv-desktop:~/test1$ ./my_mog2_test
terminate called after throwing an instance of 'cv::Exception'
  what():  OpenCV(4.1.0) /home/nv/compile/opencv-4.1.0/opencv_contrib-4.1.0/modules/cudabgsegm/src/cuda/mog2.cu:122: error: (-217:Gpu API call) invalid device symbol in function 'loadConstants'

./run.sh: line 2:  8095 Aborted                 (core dumped) ./mybin /usr/src/tegra_multimedia_api/data/Video/sample_outdoor_car_1080p_10fps.h264 1

i test it on jetson tx2 and it works normal, but failed at jetson nano, can give some tips?
thanks

2

6.2 is for TX2. For Nano (or TX1) you would use 5.3 instead. Have a look to this script.

3

mmapi has a sample with reading to cuda from camera that works. It uses v4l2 though.

4

hi, Honey_Patouceul.

thank you very much, i change it to 5.3, and all works well.

please close the topic.