Ask Your Question
0

roboearth: fails to detect object when receiving point cloud

asked 2012-03-16 02:16:26 -0500

Gemini Ma gravatar image

updated 2016-10-24 09:02:18 -0500

ngrennan gravatar image

Hi,

I'm using ros-electric-turtlebot running the roboearth.

After building up a 3D model, I use the re_kinect_object_detector to detect the object. The first time the re_kinect_object_detector node received point cloud, there's log message: (I opened some log messages in matchAspects function)

[ INFO] [1331729328.968390255]: received point cloud
[ INFO] [1331729329.068954900]: matching model: PointCloud2.kinectZoomPointCloud2
[ INFO] [1331729329.069546826]: found 2 correspondences.
[ INFO] [1331729329.069591011]: found 2 3D correspondences.
[ INFO] [1331729329.069626568]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.070866657]: found 7 correspondences.
[ INFO] [1331729329.070905464]: found 7 3D correspondences.
[ INFO] [1331729329.070943400]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.070972396]: found 0 correspondences.
[ INFO] [1331729329.071004012]: found 0 3D correspondences.
[ INFO] [1331729329.071033772]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.071294529]: found 0 correspondences.
[ INFO] [1331729329.071322469]: found 0 3D correspondences.
[ INFO] [1331729329.071354157]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.071436934]: found 0 correspondences.
[ INFO] [1331729329.071464717]: found 0 3D correspondences.
[ INFO] [1331729329.071495055]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.072120036]: found 3 correspondences.
[ INFO] [1331729329.072158913]: found 3 3D correspondences.
[ INFO] [1331729329.072194388]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.072599001]: found 0 correspondences.
[ INFO] [1331729329.072637274]: found 0 3D correspondences.
[ INFO] [1331729329.072672226]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.072792093]: found 0 correspondences.
[ INFO] [1331729329.072820856]: found 0 3D correspondences.
[ INFO] [1331729329.072856909]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.073550326]: found 2 correspondences.
[ INFO] [1331729329.073588965]: found 2 3D correspondences.
[ INFO] [1331729329.073622813]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.074955371]: found 9 correspondences.
[ INFO] [1331729329.074985202]: found 9 3D correspondences.
[ INFO] [1331729329.075034665]: matchAspects: found 3 correspondences.
[ INFO] [1331729329.075088732]: HIT 9, minerror: 1.000000
[ INFO] [1331729329.075467930]: found 1 correspondences.
[ INFO] [1331729329.075496754]: found 1 3D correspondences.
[ INFO] [1331729329.075527096]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.077067249]: found 13 correspondences.
[ INFO] [1331729329.077095463]: found 13 3D correspondences.
[ INFO] [1331729329.077138026]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.077487096]: found 1 correspondences.
[ INFO] [1331729329.077515891]: found 1 3D correspondences.
[ INFO] [1331729329.077545470]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.078075524]: found 2 correspondences.
[ INFO] [1331729329.078103767]: found 2 3D correspondences.
[ INFO] [1331729329.078134121]: matchAspects: found 0 correspondences.
[ INFO] [1331729329.078167414]: displaying 307200 points in frame: camera_rgb_optical_frame

And a image window opens, i can see the highlighted object.

But afterwards, when the node receives another point cloud:

[ INFO] [1331729341.966271561]: received point cloud
[ INFO] [1331729342.066705365]: matching model: PointCloud2.kinectZoomPointCloud2
[ INFO] [1331729342.067266809]: found 4 correspondences.
[ INFO] [1331729342.067307929]: found 4 3D correspondences.
[ INFO] [1331729342.067344704]: matchAspects: found 0 correspondences.
[ INFO] [1331729342.068547286]: found 5 correspondences.
[ INFO] [1331729342.068586474]: found 5 3D correspondences.
[ INFO] [1331729342.068620191]: matchAspects: found 0 correspondences.
[ INFO] [1331729342.068651428]: found 0 correspondences.
[ INFO] [1331729342.068683878]: found 0 3D correspondences.
[ INFO] [1331729342.068715168]: matchAspects: found 0 correspondences.
[ INFO] [1331729342.068979946]: found 1 correspondences.
[ INFO] [1331729342.069009116]: found 1 3D correspondences.
[ INFO] [1331729342.069053959]: matchAspects: found 0 correspondences.
[ INFO] [1331729342.069138500]: found 0 correspondences.
[ INFO] [1331729342.069166477]: found 0 3D correspondences.
[ INFO] [1331729342.069197986]: matchAspects: found 0 correspondences.
[ INFO] [1331729342.069836880]: found 2 correspondences.
[ INFO] [1331729342.069874740]: found 2 3D correspondences.
[ INFO ...
(more)
edit retag flag offensive close merge delete

Comments

Which image do you refer to? The one published under 're_kinect/detection_results'? Alternatively, you can try to enable DEBUG_VIS in src/recognitionmodel.h, ~line 42.

ddimarco gravatar image ddimarco  ( 2012-03-19 06:00:47 -0500 )edit

I enabled DEBUG_VIS and can see the match. but after a while, sometimes it lost the highlight points. Also, if i lay down the object that was modeled while standing up, the algorithm can't detect it either...

Gemini Ma gravatar image Gemini Ma  ( 2012-03-22 03:36:35 -0500 )edit

The methods does not guarantee 100% stability. I don't think I can easily help you there, sorry.

ddimarco gravatar image ddimarco  ( 2012-03-27 01:27:06 -0500 )edit

1 Answer

Sort by ยป oldest newest most voted
1

answered 2012-07-10 06:27:46 -0500

AksatShah gravatar image

I believe I was getting similar results to this when using RoboEarth. Fortunately for me I had to simulate my detection on a light flooring on which it worked very well. Sometimes changes in the background when performing training can cause some issues so when I performed the training I also used a light plain background. Finally I used a bounding box almost perfect to the object rather than the default amount. This means you won't get any issues with other fluctuations around the training surface.

Like ddimarco said this method does not guarantee 100% stability so don't expect a 100% detection rate. It will definitely fluctuate and I also tried it with the object lying down and unfortunately it did not work.

edit flag offensive delete link more

Your Answer

Please start posting anonymously - your entry will be published after you log in or create a new account.

Add Answer

Question Tools

Stats

Asked: 2012-03-16 02:16:26 -0500

Seen: 297 times

Last updated: Jul 10 '12