search

spencer-project / spencer_people_tracking

Multi-modal ROS-based people detection and tracking framework for mobile robots developed within the context of the EU FP7 project SPENCER.
http://www.spencer.eu/
656 stars 326 forks source link

Intergrate other detection algorithm. #85

Open lff12940 opened 3 years ago

lff12940 commented 3 years ago

I integrated other detection algorithms and used the provided scripts for message conversion, but I encountered the following problems:

[ WARN] [1608023759.960176417, 1439557604.073015311]: Jump back in time detected. Deleting all existing tracks. Old time was 1439557603.966, new time is 1439557603.945 (dt=-0.021 sec)! [ WARN] [1608023760.593699488, 1439557604.198680538]: Jump back in time detected. Deleting all existing tracks. Old time was 1439557604.099, new time is 1439557604.057 (dt=-0.042 sec)! [ WARN] [1608023761.217882125, 1439557604.323381596]: Jump back in time detected. Deleting all existing tracks. Old time was 1439557604.233, new time is 1439557604.138 (dt=-0.095 sec)!

The detection algorithm is: DROW and DR-SPAAM person detectorshttps://github.com/VisualComputingInstitute/DR-SPAAM-Detector

Can anyone help me? How to solve this problem?

THanks.

tlind commented 3 years ago

Hi, it seems that your messages arrive slightly out-of-order. An easy workaround would be to add a small tolerance in delta time in the condition of the if statement within the tracker: https://github.com/spencer-project/spencer_people_tracking/blob/09b256ba4bc22c5cae8a5ae88960de1a387cfd7f/tracking/people/srl_nearest_neighbor_tracker/src/srl_nearest_neighbor_tracker/nearest_neighbor_tracker.cpp#L227-L230

Are you running multiple detectors in parallel (all publishing to the detected persons topic that is input to the tracker), and they are not temporally synchronized?

lff12940 commented 3 years ago

Thanks. It worked when I add a tolerance in delta time.

I replaced the laser detectors with DR-SPAAM-Detector in the launch file tracking_on_bagfile.launch. At present, I have integrated DR-SPAAM detector and upper-body detector using front_laser and front_rgbd, but the fused detections are not good. There are many false detections in the test results. How should I fine tune for better result?

Thanks.

tlind commented 3 years ago

Which detector is mostly causing these false alarms? Or are the detections from the individual detectors okay, and the problem is at the fusion stage? Maybe you could attach a picture or short video.