add rmf_human_detector

[PranayShirodkar]

rmf_human_detector

ROS 2 node that subscribes to sensor_msgs::Image messages published by a monocular camera and publishes the detections over /rmf_obstacles as rmf_obstacle_msgs::Obstacles message. The node runs YOLOv5 to detect the presence of humans as obstacles.

Launch human detector

First launch rmf demos hotel world with a camera in it from this PR. Then:

ros2 launch rmf_human_detector human_detector_launch.py

[PranayShirodkar]

This _config->visualize variable is used to control if the user wants to see the final image with bounding boxes in a seperate OpenCV window. Either way, the final image is being published as a ros topic. The OpenCV window was useful during debugging to see what was going on.

Do we want to just do away with this option and not create the OpenCV window at all since it can be seen from the ros topic?

[Yadunund]

High level comments

• Let's not commit the onnx model here. It will inflate the size of the repo and more importantly we shouldn't commit binaries as they dont diff. Best option would be to provide instructions on how to download the model (could be a script) and then have the path to the model be a ROS 2 param.
• The camera mesh should also be removed from here. Either it should go on fuel or rmf_demos_assets. The latter would be easier.
• Remove the worlds. Update one of the rmf_demos_worlds to include the camera model
• Use ROS 2 params for other configurable parameters (topic names, strings, etc)
• Uncrustify
• Update README

[Yadunund]

Why do we need this camera pose? I don't understand the logic

If your code is waiting for this value to be received, it should be an std::optional<Transform>. Then if nullopt don't process.

[ahcorde]

Closing this PR in favour of this other one #3