strands-project / strands_perception_people

long-term detection, tracking and recognition of people
96 stars 70 forks source link

supporting ROS type `people_msgs/People` #143

Closed marc-hanheide closed 9 years ago

marc-hanheide commented 9 years ago

It would be very beneficial, e.g. to use the social costmaps of Lu!! et al to have our people tracker publishing on the /people topic (type: people_msgs/People):

std_msgs/Header header
  uint32 seq
  time stamp
  string frame_id
people_msgs/Person[] people
  string name
  geometry_msgs/Point position
    float64 x
    float64 y
    float64 z
  geometry_msgs/Point velocity
    float64 x
    float64 y
    float64 z
  float64 reliability
  string[] tagnames
  string[] tags

So, my question is: Is this easily possible? It seems to me it should be, any reasons why we haven't used this before? @ferdianjovan @cdondrup? I assign to @cdondrup for now to get an answer. Doesn't mean I expect you to implement this.

cdondrup commented 9 years ago

Should be easy enough to do. The only thing I can't really figure out is, what name, tags, tagnames, and reliability is used for. A quick grep on the navigation_layers repo told me that all these values are never used. I'll set name to the uuid and will also add a tag for it. The position and velocity comes out of the tracker anyway and since my code is always fail safe I will just set the reliability to 1.0 ;)

marc-hanheide commented 9 years ago

Great. We can then have the student do some experiment with that proxemics layer.

ferdianjovan commented 9 years ago

Ah yes, that one, I came across that a long time ago and I forgot about it. About the velocity, what will you put there? Will it be the velocity between two poses or be the average velocity?

cdondrup commented 9 years ago

It will be the velocity used by the tracker for the prediction. Since it uses a constant velocity model, I assume it'll be the average velocity. I didn't check the actual implementation yet, though.

cdondrup commented 9 years ago

Done. Needs testing though.

marc-hanheide commented 9 years ago

Thanks!