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/
660 stars 327 forks source link

Mismatch between range of laser detection and rgbd detection? #43

Open kajep09 opened 7 years ago

kajep09 commented 7 years ago

I have created a setup similar to what is described for the "tracking_on_robot" but with only front facing rgbd (ASUS Xtion Pro) and laser scanner. When I am close to the camera/laser scanner there is a good alignment between the laser detections and the rgbd detections and therefore the two get matched and it tracks me well. When I move further away from the camera, there is a mismatch in alignment between what range the laser scanner detects me at and what range the rgbd sensor detects me at, so it actually creates two matches even though there is only one person in the view. Any idea what could cause this mismatch or how to tune it?

Video example: https://youtu.be/ADyc6U4wN8c

tlind commented 7 years ago

Did you check the "tracking_on_bagfile" launch file? That one replicates exactly the setup you mention (1 Asus + 1 laser scanner).

Are you sure the RGB-D sensor is calibrated properly? I remember from the rqt_reconfigure GUI, that there is a parameter somewhere in the rgbd driver that affects the z scaling. We had to tune it on some of our devices to make the depth from the Asus and the 2D laser echoes align properly. (I'm assuming that the laser scanner is giving accurate measurements.)