people_tracker and leg_detector use different message types

[cdondrup]

Trying to run the people_tracker together with leg_detector yields the following error:

[ERROR] [1407405500.928511110]: Client [/people_tracker] wants topic /people_tracker_measurements to have datatype/md5sum [people_msgs/PositionMeasurement/54fa938b4ec28728e01575b79eb0ec7c], but our version has [people_msgs/PositionMeasurementArray/59c860d40aa739ec920eb3ad24ae019e]. Dropping connection.

[DLu]

Can you post more information about your configuration? Which people_tracker are you using?

[cdondrup]

I realise that was a bit too generic, sorry.

What I'm running is: roslaunch people_tracking_filter filter.launch which starts: <node pkg="people_tracking_filter" type="people_tracker" name="people_tracker" output="screen"/>

Once you start the leg_detector (leg_detector.launch or filtered_leg_detector.launch) it gives above error message that the topic types don't match.

So basically this is the created and published message but this is the message type the tracker subscribes to.

[DLu]

So this is a dumb question given that I'm the maintainer, but what does the people_tracking_filter node do?

[cdondrup]

On the contrary, that is a very good question indeed. I just tried to run it and see what is does because people_tracker sounded like what I wanted.
Long story short, no clue what it does but since it is in the official hydro repos I thought I just report the error.

[DLu]

That's fair enough. For me, people_tracking_filter is used as a library within the leg_detector, which is why it is included in the repo. The launch files in the leg_detector and people_velocity_tracker might be more useful for you.

[mkylmz]

I wonder if there's a solution now. I've been searching for quite some time now, yet I could not find a solution.

[yumdiddlyscrumptious]

Has anyone found a way around this bug?

I tried changing the input to peopleCallBack to PositionMeasurementArray and then implemented a for loop to iterate through all the people in the array, but was unable to get it to work.

If anyone has found a solution, or made their own, please let me know.

[cdondrup]

I have created a tracking framework that uses the leg detector: https://github.com/strands-project/strands_perception_people

It also has a RGB-D based upper body detector that can be used but you don't have to. Have a look at this package for how to use the tracker with any kind of detector. One of the default detectors used is the leg_detector.

[yumdiddlyscrumptious]

Hey cdondrup,

I appreciate your quick reply.

It sounds like this will be a very useful package, thanks so much!

[yumdiddlyscrumptious]

Hey @cdondrup,

After trying to use this package with my robot I have been unable to get the upper body detector to work.

I have been trying to use this package with a Turtlebot2 robot, which does not have a pan-tilt unit. I think that this is the reason for it not working, because the upper_body detection is dependent on the ground_plane node, which is subscribed to the /ptu/state topic which is not bing published.

I saw that there was a "really_fixed" version of the ground_plane_fixed.launch put forth in @lucasb-eyer 's code, but I was unable to get this working.

I was wondering, do you know of any other way to get the upper body detector to work on a robot without a ptu?

I apologize that this is now in a tangent from the original thread.

[lucasb-eyer]

This thread is not about the upper body detector at all, you're at the wrong place. If the lack of PTU for the ground plane is your problem, yes, my "really_fixed" version is the solution.

[CAI23sbP]

@cdondrup I solved this issue see a code and details
#106