Home
build details

Show: section status errors & todos local changes recent changes last change in-page changes feedback controls

Demo unsupervised image segmentation

Modified 2018-11-22 by Valentina Cavinato

This demo shows implemenation of k-means algorithm on the duckiebot for image segmentation.

In order to complete the procedure, you need your duckiebot in configuration DB18 until unit B-11 with its camera calibrated.

The duckiebot in configuration DB18 section B-11

Camera calibration completed.

Pre-flight checklist

Modified 2018-11-22 by Valentina Cavinato

Check: the duckiebot has sufficient battery

Check: the camera is calibrated

Demo instructions

Modified 2018-11-22 by Valentina Cavinato

Step 1: Now we need to run the docker container to be able to launch the camera_kmeans node on our duckiebot. Check if DOCKER_HOST variable is set

laptop $ echo $DOCKER_HOST

in case it is not, then set it

laptop $ export DOCKER_HOST=ROBOT_NAME.local

Now, run the docker container kmeans on your duckiebot

laptop $ docker -H ROBOT_NAME.local run -it --memory="800m" --memory-swap="1.8g" --net host --privileged -v /data:/data --name kmeans duckietown/devel-kmeans-unsup:master18 /bin/bash

Step 2: Launch the camera_kmeans launch file

duckiebot $ roslaunch pi_camera camera_kmean.launch veh:=ROBOT_NAME k:=k reduction:=reduction

k is number of centroids of the k-means algorithm with default of 3
reduction is size reduction factor by downsampling and it’s default is 4

We suggest to open camera_kmeans.launch file and look at the flow of data in the node file. camera_kmean.launch file is located under /catkin_ws/src/05-teleop/pi_camera/launch

Step 3:

laptop $ dts start_gui_tools ROBOT_NAME
root@user $ rqt_image_view

This will start a rqt terminal, where you can see the image output after the k-means algorithm.

Troubleshooting

Modified 2018-11-22 by Valentina Cavinato

Cannot see image in the rqt.

Check that you are receiving the compressed image using rostopic list on the duckiebot.

Image is too slow

Thats because k-means is computationally intensive. Increase the reduction parameter and check if you can run it faster

Because of mathjax bug

No questions found. You can ask a question on the website.