Fall Detection - Provide infer method that works with either image or pose

[tsampazk]

Current situation

Edit: Main issue is fixed see comment below.

Right now, the standalone fall detection tool uses a pose estimation learner internally (which is provided in the learners ctor). The inference method requires an image. Inference also returns the pose for further usage, which might be counterintuitive.

New feature

It would be helpful to be able to run fall detection inference with provided OpenDR pose instead of image, by running a pose estimator externally. This usage is closer to the described usage of "cheap fall detection when pose estimation is already performed for other purposes" and can "disentangle" the used pose estimation learner from the fall detection learner, if the user chooses.

Implementation

Out of various options (e.g. multiple infer methods with different names) and because overloading is not possible in a straightforward manner in python, i would suggest putting image and pose both as optional arguments with None as default for the existing infer method and managing them internally in the method.

ROS implications

This can also improve the fall detection ROS node, giving the option to run standalone and take the pose from a ROS message. Right now in ROS, the pose gets consumed and cannot be used for other purposes as it is not published by the node.

Any suggestions are welcome.

[tsampazk]

The main part of this issue can be considered fixed by #397.

However, to be fully fixed the fall detection ROS nodes should be able to subscribe to a pose annotation topic and work with that as input instead of the current situation where they can only subscribe to an image topic.

This way they will be able to be chained with an already running pose estimator node.

[tsampazk]

Fixed by #423