Group Tracking

This page contains additional files for our publication on group tracking:

B. Lau, K.O. Arras and W. Burgard
Multi-Model Hypothesis Group Tracking and Group Size Estimation
International Journal of Social Robotics 2 (1), 2010, pp. 19-30, Springer Netherlands
The final publication is available at www.springerlink.com. In case you don't have access, here is a free copy (PDF).

Files available for download

Log files (recorded data)
corridor.log.gz371KB
dataset1.log.gz1077KB   includes ground truth
dataset2.log.gz2511KB   includes ground truth
Map files
fr101.map.gz52KB   for dataset*.log
fr079-sm.map.gz58KB   for corridor.log
Video
grouptracker.mpeg14 MB

Explanations

The logs and the maps are stored in the formats used by the Carnegie Mellon Robotics Toolkit CARMEN. From these log files, our tracker uses the lines starting with ROBOTLASER1, i.e., laser range readings combined with odometry information. Additionally, the GLOBALPOS lines are read, which contain the global pose estimated by localization.

The ground truth (if available) is also stored in these log files. The corresponding lines start with the tag GTLABEL, and contain manually annotated information for the subsequent ROBOTLASER1 line. It contains foreground/background labels, tracks of individual people, and group tracks. The syntax is defined as follows:

GTLABEL <unlabeled> <people> <background> <occludedPeople> <peopleLinks> <groups> -1 timestamp hostname timestamp
<unlabeled>list of unlabeled clusters. In a completely labeled log file, this is just a '0'.
<people>number of people followed by that number of of people clusters
<background>number of background clusters followed by that number of background clusters
<occludedPeople>number of occluded people followed by that number of people clusters
<peopleLinks>number of people links followed by that number of link numbers
<groups>number of groups followed by that number of groups
clusternumber of points follwed by x/y coordinate pairs (global coordinates)
link numberThese numbers correspond to the IDs of people and occluded people in the last frame.
If the person was outside the sensor field of view in the previous frame, the link number is -1. The IDs are implicitly given: they correspond to the position of a person in <people> for IDs from 0 to numPeople-1, and to the position of a person in <occludedPeople> for IDs from numPeople on.
groupnumber of people in the group followed by their ID (see last line)