Compiling object_recognition against which ROS base installation?

Question

Until now I have successfully compiled object_recognition against ROS diamondback installation using the following .rosinstall configuration:

- svn: {local-name: object_recognition, uri: 'https://code.ros.org/svn/wg-ros-pkg/branches/trunk_diamondback/stacks/object_recognition'}
- svn: {local-name: object_recognition_experimental, uri: 'https://code.ros.org/svn/wg-ros-pkg/branches/trunk_diamondback/stacks/object_recognition_experimental'}
- svn: {local-name: perception_pcl_addons, uri: 'https://code.ros.org/svn/ros-pkg/stacks/perception_pcl_addons/trunk'}
- svn: {local-name: vslam, uri: 'https://code.ros.org/svn/ros-pkg/stacks/vslam/trunk'}
- svn: {local-name: opencv_candidate, uri: 'https://code.ros.org/svn/wg-ros-pkg/branches/trunk_diamondback/stacks/vision_bleeding/opencv_candidate'}
- other: {local-name: /opt/ros/diamondback/ros}
- other: {local-name: /opt/ros/diamondback/stacks}

where I installed diamondback as a .deb package in Ubuntu, following these instructions. After a SVN update of object_recognition and its dependencies (except ROS Diamondback) I am running into errors when using recognizer in tod_detecting. I receive "OpenCV errors" (some assertions fail).

• I wonder whether I am compiling the object_recognition stack against the same ROS installation and OpenCV version as the object_recognition developers?
• Or should I compile against SVN version of ROS?
• Is there a .rosinstall file for the object_recognition stack that relieves me of the burden to find the right dependencies for making object_recognition?

It would be helpful to have a similar setup like the object_recognition stack developers.

Background:

([ INFO] [1300976281.637313973]: LSH matching :1.66269
OpenCV Error: Assertion failed (CV_IS_MAT(objectPoints) && CV_IS_MAT(imagePoints) && CV_IS_MAT(A) && CV_IS_MAT(rvec) && CV_IS_MAT(tvec)) in cvFindExtrinsicCameraParams2, file /tmp/buildd/ros-diamondback-vision-opencv-1.4.1/debian/ros-diamondback-vision-opencv/opt/ros/diamondback/stacks/vision_opencv/opencv2/build/opencv-svn/modules/calib3d/src/calibration.cpp, line 1156)

I'd like to know whether this is regression or whether I just need to adapt my environment.

---

Originally posted by Julius on ROS Answers with karma: 960 on 2011-03-24

Post score: 1

---

Original comments

Comment by Julius on 2011-03-28:
See also https://code.ros.org/trac/wg-ros-pkg/ticket/5083

Comment by Julius on 2011-03-28:
It seems that the problem is not with the wrong OpenCV version I used, as both the recent SVN version from OpenCV and Diamondback's lead to failure.

Answer 1

It's a regression, revert to the version that worked for you: the code was working because it was assuming that the camera parameters were the training ones if none were given.

This was a dangerous assumption so i removed it. I did not notice it as the Kinect is giving us the data and I am working with the Kinect mostly. There is no way to input those parameters through the test file for now. We are cleaning a lot of stuff these days, and we are trying to clean the parameter files too so that should be done shortly I hope.

We are not really improving performance these days, just cleaning, so there is no point in svn up this week or next week.

Regards, Vincent

---

Originally posted by Vincent Rabaud with karma: 1111 on 2011-03-28

This answer was ACCEPTED on the original site

Post score: 1

---

Original comments

Comment by Julius on 2011-04-11:
I found two of the related bugs, see https://code.ros.org/trac/wg-ros-pkg/ticket/5083 for a quick&dirty workaround.

Comment by Julius on 2011-04-11:
I'd really like to SVN update to get the refactored version of the GuessGenerator, but it's two weeks gone and it still is an issue. I guess the camera matrix is not initialized. I'll post another comment if I get anywhere.

Comment by Julius on 2011-03-28:
Thanks for confirmation. Regression introduced with revision 50321. But after all, except for the cleanups, looking at the code in the recent revisions, I got the impression that 3d to 3d matching has been implemented in GuessGenerator.