EuRoC Forum

[ghepard1]

hi,

After some testing it seems that the disparity image obtained from the stereo camera is almost empty and it does not provide any useful information inside the faces of the workpiece where the holes are located.

We think that this is caused by the lack of textures on the workpiece.

Does anybody experienced this issue?
Would it be possible to add some textures to the workpiece to enhance the disparity?

thanks

[michieletto]

Hi ghepard1,
we made a quick test about it, even if we are not actually working on this issue right now.

Everything works fine for us.
In the attached figure you can find the point cloud visualization from rviz, the overall perspective of the robot position from kinect1 and the disparity image.

Maybe you can change your SAD window size in order to avoid ambiguities in matching points between the left and right images.

Best.
Stefano

[AIS]

Hello,

I have a doubt around the access to the stereo camera. The technical annex points that for task 2 the stereo camera should be accessed through the topic 'stereo_camera/*'. Even so, through this topic only one of the images is available (the one related with the left camera).

Is it possible to use the information from the topic '/simulation_ur5/multisense_sl/camera/*' to obtain both the left and right images of the stereo camera?

Regards

[Fabjan]

Hello,

I tried to test the stereo processing in a position closer to the workpiece and this is the result.

euroc.png (62.92 KiB) Viewed 635 times

The result does not change by varying the parameters of the algorithm. For the stereo processing we need to add a texture on the mesh of the workpiece model, otherwise the stereo sensor mounted on the arm is useless, especially given the availability of two kinect sensor in the simulation.

[admin]

Dear Fabian,

regarding texture on the object: We do not intend to add texture. The stereo cameras can be used to improve accuracy of perception especially around the holes (wrt the kinect data). As your image shows: this is where you got some data. Maybe the standard block matcher is not the best approach.

Best

[admin]

Hello,

I have a doubt around the access to the stereo camera. The technical annex points that for task 2 the stereo camera should be accessed through the topic 'stereo_camera/*'. Even so, through this topic only one of the images is available (the one related with the left camera).

Is it possible to use the information from the topic '/simulation_ur5/multisense_sl/camera/*' to obtain both the left and right images of the stereo camera?

Regards

Hi AIS,

Actually both stereo cameras should be accessible and deliver data below the topic /stereo_camera/...

Maybe there is something wrong in your setup.

Could you post the output of rostopic list?
And maybe run rostopic info for all suptopics of /stereo_camera/ and post the result as well.

In the meanwhile you can rely on /simulation_ur5/multisense_sl etc. However this topic will probably not be available during final evaluation.

Best

[michieletto]

Hello,

I have a doubt around the access to the stereo camera. The technical annex points that for task 2 the stereo camera should be accessed through the topic 'stereo_camera/*'. Even so, through this topic only one of the images is available (the one related with the left camera).

Is it possible to use the information from the topic '/simulation_ur5/multisense_sl/camera/*' to obtain both the left and right images of the stereo camera?

Regards

Hi AIS,

Actually both stereo cameras should be accessible and deliver data below the topic /stereo_camera/...

Maybe there is something wrong in your setup.

Could you post the output of rostopic list?
And maybe run rostopic info for all suptopics of /stereo_camera/ and post the result as well.

In the meanwhile you can rely on /simulation_ur5/multisense_sl etc. However this topic will probably not be available during final evaluation.

Best

Actually, as noticed also by Michal in a different topic here (Task 2: Weird stereo camera topics and TFs), the only information available in /stereo_camera are:

Code: Select all

/stereo_camera/camera_info
/stereo_camera/image_raw

maybe something is wrong with the configuration, anyway right now we are using the alternative topic /simulation_ur5/multisense_sl/camera/ from which we can access to:

Code: Select all

/simulation_ur5/multisense_sl/camera/left/camera_info
/simulation_ur5/multisense_sl/camera/left/image_raw
/simulation_ur5/multisense_sl/camera/left/image_raw/theora
/simulation_ur5/multisense_sl/camera/left/image_raw/theora/parameter_descriptions
/simulation_ur5/multisense_sl/camera/left/image_raw/theora/parameter_updates
/simulation_ur5/multisense_sl/camera/left/parameter_descriptions
/simulation_ur5/multisense_sl/camera/left/parameter_updates
/simulation_ur5/multisense_sl/camera/right/camera_info
/simulation_ur5/multisense_sl/camera/right/image_raw
/simulation_ur5/multisense_sl/camera/right/parameter_descriptions
/simulation_ur5/multisense_sl/camera/right/parameter_updates

Just ask if you need any other data.

Thank you.
Stefano

[but-spanel]

Hi all,

The listing of topics on my computer is exactly the same as Stefano posted. And all my colleagues report the same issue. It definetely seems to be a problem with configuration. Please could anybody from the EuRoC team check and clarify this? Thanks.

Btw, we have foudn another interesting issue that strongly affects how the workpiece and the holes look in the stereo camera. The color of the workpiece, i.e. result of the simulation, is different when running on different computers having the same version of Ubuntu, ROS, etc. Maybe it is related to a concrete GPU running the simulation and/or default rendering/lights settings...

Simulation output on PC1:

pc1.jpg (117.77 KiB) Viewed 591 times

Simulation output on PC2:

pc2.jpg (89.52 KiB) Viewed 591 times

Please note that images presented in the technical annex are more similar to the "darker ones" where a detection of the holes might be very difficult as one of them is almost missing being unrecognizable...

Best,
Michal

[admin]

Dear all,

regarding the mixed topics for the stereo camera. There was a discrepancy in the technical annex.
Actually the the image data and the calibration data to be used for task 2 is located under /simulation_ur5/multisense_sl/

The topic /stereo_camera/ is not valid (contrary to what was stated in the tech annex on page 13 regarding task 2) and will not be available in the final evaluation.

Best Regards