Update leg_detector to OpenCV3 by kdhansen · Pull Request #53 · wg-perception/people

kdhansen · 2017-03-28T20:25:10Z

I have updated the leg_detector to the new structure of OpenCV3. I actually had another PR #48 open but it failed in Travis. Isaac Saito fixed this in his PR #49 which I rebased this code onto, and now it passes on Travis!

Isaac also found that we need to hardcode the location of haarcascade_frontal_face_alt from OpenCV3 (see #52). Between the time he made #49 and now, OpenCV3 was updated to 3.2.0, so I went and updated the filepath.

This is all to say that, merging this branch also includes all the hard work of Isaac in PR #49.

…V 3.1.0.

[face_detector][test common] Fix warning in .launch format. Fix namespace in testcases. Increase retry. .bag-based perceptional tests are hard to deterministic, so it makes sense to do this. Set param use_sim_time. [face_detector][test] hztest can be used for measureing no publishment. hzerror is not needed in that case. [face_detector][test] Specify test name. Pass all args. [face_detector][test] pass_all_args might require all args kept out of include tag. [face_detector][test] Specify test-name. [face_detector][test] Disable all tests other than rgbd. I couldn't figure out how to use launches other than face_detector.rgbdlaunch, after hours of trial on local machine. I leave that up to someone else. [face_detector][test] Detection using rgbd works with this config. [face_detector][test] Shorter test duration (bag files are less than 7 seconds). [face_detector][test] Shorter test_duration. We only need a second or so as long as we can confirm the designated topic is being published. [face_detector][test] rosbag play with clock and loop option.

Since the publish rate of the detected face varies every second because one of the persons in the bag file is moving, [publishtest](http://wiki.ros.org/rostest/Nodes#publishtest) should suit the best in this case. [face_detector][test common] Enable substitution in rosparam tag.

The cv::ml package has been rewritten. I just changed the code to accomodate the new api, maybe this does not conform to the way the new cv::ml should be used...

In opencv3 the predict_prob method was removed from cv::ml::RTrees.

The dependency on the people_msgs package should be enough. This removes warnings for the leg_detector and people_tracking_filter when building with catkin_make_isolated and catkin build.

This reverts commit 0bda254.

Isaac Saito has been working hard to make the Travis CI work with Kinetic. Removing his email from ravis.yml, to save his inbox from errors due to other issues.

Version changed to 3.2.0, unfortunately we need to hardcode the location for now.

mjyc · 2017-11-24T16:55:45Z

Hi! Is there a plan for reviewing this PR? If not, is there any way I can help? e.g. code review, comparing performance on real data between this & 14.04, etc.

awesomebytes · 2018-04-20T08:31:15Z

Bump, any plan on moving this forward?
I can try on a PR2 both the indigo and kinetic versions.

awesomebytes · 2018-04-20T09:19:22Z

So, in case it helps on deciding, I gave it a try to run it on the indigo-devel version on a Ubuntu 14.04 and on the kinetic version this PR has on a VM with Ubuntu 16.04.

Both are running with the same laser topic, a real Hokuyo UTM-30LX-01 (from a PR2).

Find a video of their behaviour in Rviz here: https://youtu.be/G-OxB-L14z0
(Left side indigo, right side kinetic).

Find also attached to this comment a rosbag with all the topics:

path:         leg_detector_indigo_vs_kinetic_2018-04-20-19-05-57.bag
version:      2.0
duration:     34.4s
start:        Apr 20 2018 19:05:58.10 (1524215158.10)
end:          Apr 20 2018 19:06:32.47 (1524215192.47)
size:         6.6 MB
messages:     77457
compression:  bz2 [40/40 chunks; 19.23%]
uncompressed: 29.7 MB @ 884.3 KB/s
compressed:    5.7 MB @ 170.1 KB/s (19.23%)
types:        people_msgs/PositionMeasurementArray [59c860d40aa739ec920eb3ad24ae019e]
              sensor_msgs/LaserScan                [90c7ef2dc6895d81024acba2ac42f369]
              visualization_msgs/Marker            [18326976df9d29249efc939e00342cde]
topics:       /base_scan                               738 msgs    : sensor_msgs/LaserScan               
              /leg_tracker_measurements                737 msgs    : people_msgs/PositionMeasurementArray
              /leg_tracker_measurements_kinetic        737 msgs    : people_msgs/PositionMeasurementArray
              /people_tracker_measurements             737 msgs    : people_msgs/PositionMeasurementArray
              /people_tracker_measurements_kinetic     737 msgs    : people_msgs/PositionMeasurementArray
              /visualization_marker                  35598 msgs    : visualization_msgs/Marker           
              /visualization_marker_kinetic          38173 msgs    : visualization_msgs/Marker

leg_detector_indigo_vs_kinetic_2018-04-20-19-05-57.bag.tar.gz

I see even an improvement in kinetic. But generally, they do quite the same job.

This was running (after checking out the repo and doing catkin_make install and sourcing that install):

roslaunch leg_detector leg_detector.launch

With the only modification of changing the fixed_frame to laser, and on the kinetic version, remapping the topics to end in kinetic.

I'd like to have this package back in the ROS repos as it's very good to enable simple demos (like this video) to get people started with ROS and stuff to do with people perception.

Thank you all for your very good work! @DLu @kdhansen

dlaz

Overall LGTM, just a few small comments.

dlaz · 2018-04-23T18:38:34Z


-  add_rostest(test/wide-stereo_true_rtest.xml)
-  add_rostest(test/wide-stereo_false_rtest.xml)
+#  add_rostest(test/wide-stereo_true_rtest.xml)


I'd prefer just deleting these instead of commenting them out, here and below

dlaz · 2018-04-23T18:40:24Z

  int mask_count_;

-  cv::ml::RTrees forest;
+  // cv::ml::RTrees forest;


nit: remove

dlaz · 2018-04-23T18:42:21Z

I don't really have things setup to try this out now, so unless there are any objections from @DLu I'm going to go ahead and merge. I'll check back in a few days.

awesomebytes · 2018-04-30T04:39:26Z

@dlaz a little friendly bump in case anyone wants to take a final look or you want to merge it.

130s and others added 11 commits February 17, 2017 16:17

[kinetic] Update Travis config to run on Ubuntu 16.04 via Docker.

52e8547

[test] Missing dependency.

e2b25b1

[face_detector] an obtuse fix to hardcoded path for Xenial with OpenC…

9d274b3

…V 3.1.0.

Fix CvMat change in OpenCV3

665d3b0

The cv::ml package has been rewritten. I just changed the code to accomodate the new api, maybe this does not conform to the way the new cv::ml should be used...

Fix missing fuzzy prediction from opencv2

d4f348e

In opencv3 the predict_prob method was removed from cv::ml::RTrees.

Remove dependency on people_msgs generation

67317e7

The dependency on the people_msgs package should be enough. This removes warnings for the leg_detector and people_tracking_filter when building with catkin_make_isolated and catkin build.

Revert "Remove no build packages from metapackage."

0ce4c42

This reverts commit 0bda254.

Save Isaac from getting emails about failed builds.

886f465

Isaac Saito has been working hard to make the Travis CI work with Kinetic. Removing his email from ravis.yml, to save his inbox from errors due to other issues.

Update hardcoded location of opencv

2e6c18b

Version changed to 3.2.0, unfortunately we need to hardcode the location for now.

DLu added the legs label Apr 9, 2017

dlaz requested a review from DLu December 20, 2017 16:53

awesomebytes mentioned this pull request Apr 22, 2018

rosdistro sync, Wed Apr 18 21:29:45 2018 ros/ros-overlay#571

Merged

56 tasks

dlaz approved these changes Apr 23, 2018

View reviewed changes

dlaz merged commit 20257a4 into wg-perception:kinetic Apr 30, 2018

DLu mentioned this pull request Aug 28, 2018

leg_detector kinetic rtree issue #44

Closed

semeyerz mentioned this pull request Apr 1, 2019

Random Leg Probabilities due to Incorrect Copy of Features #67

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Update leg_detector to OpenCV3 - #53

Update leg_detector to OpenCV3#53
dlaz merged 11 commits into
wg-perception:kineticfrom
kdhansen:kinetic

kdhansen commented Mar 28, 2017

Uh oh!

mjyc commented Nov 24, 2017

Uh oh!

awesomebytes commented Apr 20, 2018

Uh oh!

awesomebytes commented Apr 20, 2018 •

edited

Loading

Uh oh!

dlaz left a comment

Uh oh!

dlaz Apr 23, 2018

Uh oh!

dlaz Apr 23, 2018

Uh oh!

dlaz commented Apr 23, 2018

Uh oh!

awesomebytes commented Apr 30, 2018

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

Uh oh!

Conversation

kdhansen commented Mar 28, 2017

Uh oh!

mjyc commented Nov 24, 2017

Uh oh!

awesomebytes commented Apr 20, 2018

Uh oh!

awesomebytes commented Apr 20, 2018 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dlaz left a comment

Choose a reason for hiding this comment

Uh oh!

dlaz Apr 23, 2018

Choose a reason for hiding this comment

Uh oh!

dlaz Apr 23, 2018

Choose a reason for hiding this comment

Uh oh!

dlaz commented Apr 23, 2018

Uh oh!

awesomebytes commented Apr 30, 2018

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

awesomebytes commented Apr 20, 2018 •

edited

Loading