rgbd images from realsense cameras

[tajgr]

• png format for rgb and infrared images (jpeg default),
• rgbd images with help of the rs.align

note: The reason for png is that my colleagues required some lossless format.
Moreover, the Multicam support individual resolution and depth_fps now. Some types of realsense cameras required different values (D455 or L515).

[m3d]

• png format for rgb and infrared images (jpeg default),
• rgbd images with help of the rs.align

can you add some motivation? the depth encoded as PNG is much bigger as far as I remember
p.s. is it still "work in progress"=draft, or ready for review?

[tajgr]

After a long time, i returned to this PR.
@m3d Thanks for comment. I have updated the description. Now it is ready for review.

[tajgr]

This is incompatible with virtual subt logs because there are different units (mm versus m). I think the mm (int16) are better choice . Or not?

[m3d]

this is breaking change so I do not like it ... I actually do not like the RGBD bundle, but I understand it from "synchronized source" point of view. Yes it is not practical and the data are bigger (twice?)

[tajgr]

Sample log:
test-realsense-D400-220112_132945.log

$ python -m osgar.logger ../osgar_logs/test-realsense-D400-220112_132945.log
k name bytes | count | freq Hz
-----------------------------------------
0 sys 431 | 4 | 0.7Hz
1 app.rgbd_raw 6859742 | 49 | 8.9Hz
2 app.infra 0 | 0 | 0.0Hz

Total time 0:00:05.488617

[m3d]

I do not like this IF switch much - looks like different class should be used?

[tajgr]

Could be better this?
bus.register('depth:gz', 'color', 'infra', 'rgbd_raw:gz')
The IF switch will remain just in the depth_callback.

[m3d]

@jisa any comments to larger RGBD bundles? To be honest I do not really know :(.
I also do not have realsense to verify that the functionality is not broken. I see the motivation to keep synchronized RGBD but somehow I am not happy with it.

[m3d]

@tajgr - please create new "bundle", which will be depth uint16 in millimeters and RGB image (JPEG?). You can drop the camera to robot and robot absolute positions. This new structure will reflect only raw RGBD synchronized data from Realsense and can be easily converted to existing "ROS compatible RGBD bundle". OK? Thanks m.
p.s. lidarview can be then extended to support also this type of bundle

[tajgr]

Thanks. I created new bundle where dictionary is used. Yes, it is not exactly what you said.
My motivation is structure that could be more universal and more usable in the future. The bundle should always contains "data" and optionally "camera_pose" (maybe better sensor_pose) "robot_pose" or "metadata". The names should be firmly given. Is it ok for you @m3d ? If so i will complete it in Multicam.