Make can't find my opencv.hpp file - how to direct it?

Question

I have a dual camera hardware module for my raspberry pi zero. It requires some software to get working though. I have cloned the required repo from, as seen here:

pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ ls
arducam_multi_adapter_uc444.cpp  Makefile  obj  readme.md  shell

When I try to make, I get the following error:

pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ sudo make
g++ -c -o obj/arducam_multi_adapter_uc444.o arducam_multi_adapter_uc444.cpp `pkg-config --cflags --libs opencv`  
arducam_multi_adapter_uc444.cpp:4:10: fatal error: opencv2/opencv.hpp: No such file or directory
    4 | #include <opencv2/opencv.hpp>
      |          ^~~~~~~~~~~~~~~~~~~~
compilation terminated.
make: *** [Makefile:10: obj/arducam_multi_adapter_uc444.o] Error 1

Looking at /usr/include/opencv4/opencv2 I see the file this script wants.

pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ ls /usr/include/opencv4/opencv2
alphamat.hpp     ccalib           dnn_superres.hpp  freetype.hpp  imgcodecs                line_descriptor.hpp  opencv_modules.hpp    quality       shape                 superres              video          ximgproc
aruco            ccalib.hpp       dpm.hpp           fuzzy         imgcodecs.hpp            mcc                  optflow               quality.hpp   shape.hpp             superres.hpp          video.hpp      ximgproc.hpp
aruco.hpp        core             face              fuzzy.hpp     img_hash                 mcc.hpp              optflow.hpp           rapid.hpp     stereo                surface_matching      videoio        xobjdetect.hpp
bgsegm.hpp       core_detect.hpp  face.hpp          hdf           img_hash.hpp             ml                   phase_unwrapping      reg           stereo.hpp            surface_matching.hpp  videoio.hpp    xphoto
bioinspired      core.hpp         features2d        hdf.hpp       imgproc                  ml.hpp               phase_unwrapping.hpp  rgbd          stitching             text                  videostab      xphoto.hpp
bioinspired.hpp  datasets         features2d.hpp    hfs.hpp       imgproc.hpp              objdetect            photo                 rgbd.hpp      stitching.hpp         text.hpp              videostab.hpp
calib3d          dnn              flann             highgui       intensity_transform.hpp  objdetect.hpp        photo.hpp             saliency      structured_light      tracking              viz
calib3d.hpp      dnn.hpp          flann.hpp         highgui.hpp   line_descriptor          opencv.hpp           plot.hpp              saliency.hpp  structured_light.hpp  tracking.hpp          viz.hpp

I tried to create an opencv.pc as described in this earlier question, but the opencv.hpp is still not being found.

pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ cat /usr/local/lib/pkgconfig/opencv.pc 
prefix=/usr
exec_prefix=${prefix}
includedir=${prefix}/include
libdir=${exec_prefix}/lib

Name: opencv
Description: The opencv library
Version: 2.x.x
Cflags: -I${includedir}/opencv -I${includedir}/opencv2
Libs: -L${libdir} -lopencv_calib3d -lopencv_imgproc -lopencv_contrib -lopencv_legacy -lopencv_core -lopencv_ml -lopencv_features2d -lopencv_objdetect -lopencv_flann -lopencv_video -lopencv_highgui

I felt like that opencv.pc is not point to the right place as it resolves to /usr/include/opencv2 and on my pi, I have the files in /usr/include/opencv4/opencv2, though even with tinkering with where the opencv.pc file points, I'm still running into the error above. I have since returned it to its original state and have modified my .bashrc as they suggested.

pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ pkg-config --cflags --libs opencv
-I/usr/include/opencv -I/usr/include/opencv2 -lopencv_calib3d -lopencv_imgproc -lopencv_contrib -lopencv_legacy -lopencv_core -lopencv_ml -lopencv_features2d -lopencv_objdetect -lopencv_flann -lopencv_video -lopencv_highgui
pi@luna:~/RaspberryPi/Multi_Camera_Adapter/Multi_Adapter_Board_2Channel_uc444 $ echo $PKG_CONFIG_PATH
:/usr/local/lib/pkgconfig

I feel like something needs to be linked or pointed to somewhere to get make to see that opencv.hpp file. If I go in and manually edit arducam_multi_adapter_uc444.cpp's 4th line to the absolute path of the opencv.hpp, I no longer get an error in arducam_multi_adapter_uc444.cpp, but in opencv.hpp as it then is included and starts looking for core.hpp and other .hpps in /usr/include/opencv4/opencv2. How can I fix this, I feel close, I just don't know how to tell make where to look.

Answer 1

This has nothing to do with make per se. The problem is that your compiler invocation is wrong for the include statement you have.

You can debug this yourself: first, see the command make invoked:

g++ -c -o obj/arducam_multi_adapter_uc444.o arducam_multi_adapter_uc444.cpp `pkg-config --cflags --libs opencv`

Now run that command yourself from your shell prompt. You'll see it fails the same way. So you know it's not make, but rather the compiler command that's wrong.

Next, figure out what the pkg-config is expanding to; run that from your shell prompt:

pkg-config --cflags --libs opencv

It will print out the options it's adding. These options are wrong and causing your compile to fail.

First, you are compiling a source file into an object file, so you don't need the --libs option. The --libs option provides libraries to link with and these are not used during compilation.

Second, your configuration says this:

Cflags: -I${includedir}/opencv -I${includedir}/opencv2

which means the compiler will see options like -I.../include/opencv -I.../include/opencv2. There's no magic to the way header files are looked up: the compiler takes the paths provided by -I, and appends the path it needs from your #include line, one by one, until the header is found. That's all.

Since you're including opencv2/opencv.hpp, it will add that path to each of the -I paths. So, the compiler looks for .../include/opencv/opencv2/opencv.hpp first, then it will look for .../include/opencv/opencv2/opencv2/opencv.hpp.

Clearly, the header you want is not available at either of those paths: the header you want is (presumably) .../include/opencv2/opencv.hpp.

So, if you want your #include to specify opencv2/opencv.hpp, your cflags need to be:

Cflags: -I${includedir}