Add NV12 to pixel formats

[lcschaeper]

I propose this PR to enable ROS2 support of NV12 pixel format as image encoding. The NV12 pixel format is a common output format of hardware-accelerated decoders. NV12 is quite similar to the NV21 pixel format, with the exception that the order of the interleaving of U and V is reversed. The hardware-accelerated video decoder I currently use only supports NV12 as output pixel format.

NV12 corresponds to the FourCC, which was previously recommended as the identifier (see also #204 and #214).

The image_encodings.hpp file has been extended with the necessary additions.

References to show unambiguousness of the NV12 pixel format:

• V4L2: https://github.com/torvalds/linux/blob/v5.19/include/uapi/linux/videodev2.h#L599
• Kernel: https://github.com/torvalds/linux/blob/v5.19/include/uapi/drm/drm_fourcc.h#L276
• FFmpeg: https://github.com/FFmpeg/FFmpeg/blob/n5.1.2/libavcodec/raw.c#L73

[quic-zhaoyuan]

Hi, is there any merge plan for this PR, hope can use this format in Humble

[christianrauch]

I was already wondering about this for the other "NV" formats: These formats do not seem to be supported by tooling in the ROS ecosystem. You can publish images in this encoding, but none of the official ROS tools, such as cv_bridge, supports the format. Why is it beneficial to add the encoding "key" to the header files? You can already place arbitrary image data in the data field and choose an arbitrary string for the encoding. Your custom producers and consumers could agree on which encoding keys to use, e.g. NV12 in this case, and on how this format is represented, without the need to list the encoding key in the header file.

Adding these encodings to the image_encodings.hpp header does not enable support for this format in ROS 2 per se. IMHO, to "support" an encoding means that the tolling in the ecosystem, such as the cv_bridge, understands these formats. If we just add encodings like this, we could essentially also just define that the Image can hold any image data that can be represented by the 32 bit FourCC, without adding this to image_encodings.hpp and without actually supporting this with tooling.