Accuracy of AR markers tracking without depth data improvements?
Hi,
I have been using the individualMarkers node from the ar_track_alvar package successfully. With the individualMarkersNoKinect however, the detected markers are about 5 cm 'below' the right location. Apparently this depth data based improvement is important: if I comment out the
int ret = PlaneFitPoseImprovement(i, m->ros_corners_3D, selected_points, cloud, m->pose);
line in the individualMarkers code, I get a similarly large error.
Is that really the best accuracy you can get without using the depth data?
I am using a close range primesense camera with markers not more than 1 meter away from the camera. The marker size is 7 cm.
When using the individualMarkersNoKinect node, I use the following topics:
/camera/rgb/image_raw
/camera/rgb/camera_info
Thanks for any feedback,
Bert
In the meantime, I did a calibration of the rgd camera (using http://wiki.ros.org/camera_calibration) and that improved the 2d based tracking accuracy. The errors are now < 2 cm in the direction from the camera to the marker location. Depending on the camera quality and the calibration quality, I assume now that's the best possible...