Some problems about version committed on Dec. 23

[yanmingz]

Hi Mathieu,

Happy new year!

I have checked the version you committed on Dec. 23. And I have some problems.

1. In CameraStereoImages::init, the rectification function is disabled for initialization, but not enabled after initialization.

2. What is the difference between your implementation of util2d::calcOpticalFlowPyrLKStereo and cv:calcOpticalFlowPyrLK?

3. What is the version of rtabmap can work with the newest version of rtabmap_ros? I found the newest version of rtabmap_ros (committed on Dec. 18) does not support the API pcl::PointCloud<pcl::PointXYZRGB>::Ptr cloudFromStereoImages(
                const cv::Mat & imageLeft,
                const cv::Mat & imageRight,
                const StereoCameraModel & model,
                int decimation).
It still uses the parameters of cx, cy, fx, fy and baseline.

4. It seems like many settings in Windows->preference work only for loop-closure detection. Is it a good idea to separate the setting UI into 2 parts, one for loop closure detection and optimization, the other for odometry.

5.Currently the rtabmap_ros on my desktop can only show the 3D dense map of current frame, not the whole 3D dense map. I guess some settings are changed by myself unintentionally but I can't changed it back. Could you tell me which setting I should change?

6. I want to profile your program. I mean I want to know how long it takes to excute a function. I find there is a class named as Statistics which record many that kind of information. but I can't find where do you output those information. Could you tell me how to utilize the class? Or where do you save those information when the program existed?

Thanks a lot and have a nice day.

Yanming

[matlabbe]

Hi Yanming,

Happy New Year too!

1. This is a bug. Fixed in this commit.

2. The only difference is on this line (in OpenCV github, refer to this line):

Original:
- cv::Point2f delta( (float)((A12*b2 - A22*b1) * D), (float)((A12*b1 - A11*b2) * D));

Modification: delta on x-axis only, search on horizontal epipolar lines as left and right images are rectified
+ cv::Point2f delta( (float)((A12*b2 - A22*b1) * D), 0);

3. The 0.11 branch is a work-on-progress branch. Depending if I am updating the standalone or ros package, the branch may differ between standalone and ros. I will update the rtabmap_ros 0.11 branch soon. Normally, the master branch (default) should always match between the rtabmap library and rtabmap_ros.

4. In 0.11, all parameters about visual registration are now shared between odometry and loop closure constraint computation. In the code, they are under group "Vis/". In Preferences dialog, they are under RTAB-Map Settings (Advanced)->Motion Estimation->Visual Registration panel. This avoids duplicating parameters that should remain the same for consistency (odometry constraints are computed the same way as loop closure constraints).

5. You can go in Preferences->General Settings (GUI)->3D Rendering and click "Restore Defaults" at the bottom of the panel. If this doesn't help, make sure that rtabmapviz is receiving the rtabmap/mapData topic (at least the "Loop closure detection" view must show something if the topic is received).

6. There are two approaches to do that.
a. In the loop closure detection papers, I used the log mechanism (see Preferences->RTAB-Map Settings (Advanced)->Log statistics). This will create LogI.txt and LogF.txt (LogF.txt contains some timing statistics). You can open them using the showLogs.m script in MATLAB (Octave may work too). Example on Benchmark tutorial.
b. Look at the Plot statistics tutorial to view statistics live with the rtabmap interface.

cheers,
Mathieu

[yanmingz]

Hi Mathieu,

Thanks a lot.

Best regards
Yanming