We have made a lot of changes, and it is time to do a release to the buildfarm

We need a tool to determine the camera pose by observing a single fiducial in a known orientation (wall or ceiling).

• Initialize map with a fiducial

• Create fiducials to print

There is a lot of stuff that isn't used anymore

Make sure that we have all the install rules that we need to actually use pre-built packages

Describes building under indigo

We should start using C++11 concepts in fiducials to help clean up the code.

With kinetic, all supported platforms have C++11 support.

This would be useful for implementing #1

Conceptually, how is the camera pose estimated from multiple markers? Does having multiple markers improve the pose estimate?

Bag files are big, and we can get away with static images for some things

• fiducial_detect
• aruco_detect
• fiducial_slam (localization)

fiducial_detect already does this

Need to support having a subset of fiducials in a frame other than map, so that they could be placed on moving bodies.

Please consider supporting:

• robot_pose_ekf
• robot_localization

I believe support should only require a minor modification to fiducial_slam where instead of only outputting a /map tf frame you would also need to provide a new corrected odom topic.

I have been trying to use USB cameras with the ArUco detection. What I am running into with both aruco_detect and fiducial_detect is that the process dies with exit code -11. With aruco_detect it dies immediately with:

[ INFO] [1507580447.119276852]: Aruco detection ready
[aruco_detect-1] process has died [pid 229, exit code -11, cmd /opt/ros/kinetic/lib/aruco_detect/aruco_detect /camera/compressed:=/usb_cam/ima
ge/compressed /camera_info:=/usb_cam/camera_info __name:=aruco_detect __log:=/root/.ros/log/4cf040da-ad2f-11e7-abe2-0242ac110002/aruco_detect-
1.log].
log file: /root/.ros/log/4cf040da-ad2f-11e7-abe2-0242ac110002/aruco_detect-1*.log

and with fiducial_detect it loops:

process[fiducial_detect-1]: started with pid [418]
[ INFO] [1507580566.238885018]: Fiducials Localization ready
camera intrinsics:
0.000000 0.000000 0.000000 
0.000000 0.000000 0.000000 
0.000000 0.000000 0.000000 

[fiducial_detect-1] process has died [pid 418, exit code -11, cmd /opt/ros/kinetic/lib/fiducial_detect/fiducial_detect /camera/compressed:=/us
b_cam/image/compressed /camera_info:=/usb_cam/camera_info __name:=fiducial_detect __log:=/root/.ros/log/4cf040da-ad2f-11e7-abe2-0242ac110002/f
iducial_detect-1.log].
log file: /root/.ros/log/4cf040da-ad2f-11e7-abe2-0242ac110002/fiducial_detect-1*.log
[fiducial_detect-1] restarting process

and I will then see in dmesg:

[459309.981281] aruco_detect[7187]: segfault at 0 ip 000000000041ac90 sp 00007ffc833be428 error 4 in aruco_detect[400000+4b000]
[459429.206708] fiducial_detect[7399]: segfault at 0 ip 00007faa401d1770 sp 00007ffd139c34b0 error 4 in librpp_pose.so[7faa401b2000+29000]

I reproduced this on a stock Ubuntu 16.04 configuration, as well as on Docker. I tried two different USB cameras using usb_cam (I observed the same results with libuvc_camera, however, I had neglected to write down details of my test). The two cameras are:

Bus 001 Device 011: ID 041e:4097 Creative Technology, Ltd Live! Cam Chat HD [VF0700]
Bus 001 Device 012: ID 0c45:6341 Microdia Defender G-Lens 2577 HD720p Camera

To reproduce, start from the Dockerfile below:

FROM ros:kinetic-ros-base
RUN apt-get update && apt-get install -y \
    tmux \
    ros-kinetic-usb-cam \
    ros-kinetic-fiducials \
    ros-kinetic-image-view \
    && rm -rf /var/lib/apt/lists/*

I used a script something like:

xhost local:root
docker run -ti --rm -e DISPLAY=$DISPLAY --device /dev/video0 -v /tmp/.X11-unix:/tmp/.X11-unix ros_kinetic_desktop

Then run the following in separate terminals (tmux?):

roslaunch usb_cam usb_cam-test.launch
roslaunch aruco_detect aruco_detect.launch camera:=/usb_cam

Or the same with aruco_detect replaced with fiducial_detect. The first command should open a window showing images streaming in; the second should trigger the errors.

We need tests! We can start out with some basic ones.

• Publish image(s) of a fiducial in frame, and make sure that the detection outputs a sane vertices.
• Use existing map, publish image of fiducial, check for sane localization.
• Image with two fiducials, connect them
• Bag file with video of moving through a bunch of fiducials, map should make sense (Maybe not as basic)

Get rid of it, or fix it.

We got rid of old ones, now we need to add new ones

I get the following error, when I compile the package with catkin_make:

fatal error: fiducial_msgs/Fiducial.h: No such file or directory

This can be fixed by inserting

add_dependencies(rpp_pose ${${PROJECT_NAME}_EXPORTED_TARGETS} ${catkin_EXPORTED_TARGETS})

in fiducial_pose/CMakeList.txt

I am on Ubuntu 16.04, Opencv 3.2.0, ROS Kinetic

The tf pose of the detected fiducials if very spatially correct however the visualization markers of the same fiducials has high vertical error.

We need to get a working release out on Kinetic

• fiducial_slam should build a map even if base_link to camera tf is not known
• fiducial_detect should detect fiducials and publish vertices even if no camera_info

To cope better with varying lighting conditions:

• If a fiducial is not found, increase the range of thresholds that are tried
• Narrow the threshold search if fiducials are found to the range that found all of them
• If a threshold is close to 0 or 255, request a change in brightness from the image source (via dynamic configure interface or a new service)

When there are no markers, we should publish an empty fiducial_vertices

We need to get a working release out on Indigo

Very minor errors in marker angular pose estimates cause large lateral jumps in the /map frame. The 3DoF position data is great while the full 6DoF pose is extremely noisy.

In this below case I'm not sure why the tf and markers are soo far away in both X & Y lateral directions while Z looks good:

The aruco code works better than our original fiducials, but uses a different tag format.

We should be able to create a 'dictionary' for aruco that allows it to work with our fiducials.

scale of 0 in one of x/y/z

It would be useful to enable better testing.

Get rid of the marker publishing

All the travis builds are failing with some quaternion error

https://travis-ci.org/UbiquityRobotics/fiducials/builds/194241805