Robotics Class 2011/Assignment 2

Write a ROS node that subscribes to the image topic /stereo/left/image_rect (which has message type sensor_msgs/Image), and publish two topics. The first topic is a topic named "face_view" that is an image topic that has a rectangle around any face that is seen. The second topic, named "face_coords" is a PointStamped message that has point.x and point.y set to the center of the identified face if there is one face. If there is more than one face, the behavior can be implementation dependent. You can learn about the structure of the PointStamped message by typing:

rosmsg show PointStamped

The bag file distributed in class contains the image data that should be used to test your face detector. The bag file can be played back by typing:

rosbag play -l 2011-06-18-12-38-55.bag

Note that the additional "-l" argument allows the bag file to be looped indefinitely.

Once the bag file has begun playing, you can verify the image stream by viewing the raw images from the bag file by typing:

rosrun image_view image_view image:=/stereo/left/image_rect

The image viewing program is part of the image_view ROS package (which is, in turn, part of the ROS image_pipeline).

Once the image stream is verified to be working, you can begin developing your face detection system.

There is a complete example of the homework in the HacDC ROS repository. If you would like to refer to it, you can check it out via:

cd to where you store your downloaded ROS packages svn co http://hacdc-ros-pkg.googlecode.com/svn/trunk/face_detection rosmake face_detection

Note that the homework solution uses cv_bridge to map between ROS image messages and OpenCV image arrays, and it makes use of the OpenCV image blurring function cv.Smooth and face detection function cv.HaarDetectObjects.

In order to actually run the demonstration face detector, you will need to use the roslaunch tool. When you check out the above face_detection module from the HacDC repository, you will find inside the package a directory named launch. Inside this directory is a ROS launch file designed to start up the face detection module. The roslaunch system is useful for starting up a large number of nodes on your robot, instead of manually starting each node in a separate window using rosrun....

In order to start the face detection example node, you may type:

roscd face_detection roslaunch launch/face_detector.launch

As the assignment requests, the example face_detector node outputs two topics, one called "/face_view" and one called "/face_coords". You can view the "/face_view" image topic like you would any other image topic (using image_view):

rosrun image_view image_view image:=/face_view

If you wish to view the "/face_coords" topic, you made use the simple rostopic echo covered previously:

rostopic echo /face_coords

Note that the "/face_coords" topic is only transmitted (in the face detection example node at least) when faces are detected.

If you look inside the launch file, you will note that it is XML, and you can read more about the structure and meaning of these files in the roslaunch area of the ROS wiki.

For the face detection example, we simply start a node and also define a "private" parameter that is loaded to the ROS Parameter Server. As mentioned in class, this parameter is named "classifier" and it contains the filename to the Haar Cascade used by the OpenCV face detection algorithm.