I answer this so that someone stumbles upon this problem / question, no need to do the same research as me.
Apparently, access and processing of one image do not quite correspond to the definition of augmented reality. There is one more keyword for this that I already forgot (sorry), but if you want to use Google, you should not focus only on AR software.
To solve my problem, I used two approaches. The first is to use ARToolKit along with gstreamer and ffmpeg. I converted one image to one second of the video using ffmpeg, exported the line ARTOOLKIT_CONFIG and then displayed the image using ARToolKit. This was not very cool, as it is very limited, and I cannot get the reverse image without processing the output of artoolkit-opengl.
: OpenCV-Library . - : http://dasl.mem.drexel.edu/~noahKuntz/openCVTut10.html
, Irrlich-Engine ( Ogre, Irrlicht ), . Irrlicht transformDataToImage().
, X- , .
Ps: , , . .