1. This genom3 module is based on the ROS package called find_object_2d (http://wiki.ros.org/find_object_2d). It is absolutely necessary to install it before proceeding.

2. To install find_object_2d, open a new terminal and type:

   sudo apt-get install ros-indigo-find-object-2d
   

3. Clone this repository and compile the genom3 module. To do so, enter the following commands:

    genom3 skeleton -l c++ -i objectdetection.gen
    ./bootstrap.sh
    mkdir build
    cd build
    ../configure --prefix=$ROBOTPKG_BASE --with-templates=ros/server,ros/client/c
    make
    make install
   

4. Go to the directory where find_object_2d was installed:

    roscd find_object_2d/
   

5. Create a file called stereo.launch in the launch directory where find_object_2d was installed and include the following:

    <launch>
    	<!-- Nodes -->
        <group ns="find_objects_2d_Left">
    	    <node name="find_object_2d" pkg="find_object_2d" type="find_object_2d" output="screen">
    		    <remap from="image" to="/stereo/left/image_rect_color"/>
    		    <param name="gui" value="false" type="bool"/>
    		    <param name="session_path" value="/home/USER_PATH/objectdetection/sessionOBJECTS.bin" type="str"/>
    	    </node>
        </group>
    
        <group ns="find_objects_2d_Right">
    	    <node name="find_object_2d" pkg="find_object_2d" type="find_object_2d" output="screen">
    		    <remap from="image" to="/stereo/right/image_rect_color"/>
    		    <param name="gui" value="false" type="bool"/>
    		    <param name="session_path" value="/home/USER_PATH/objectdetection/sessionOBJECTS.bin" type="str"/>
    	    </node>
        </group>
    </launch>
   

6. Modify image if needed and session_path accordingly to your setup (after finishing step 8).

7. Run ros

    roscore &
   

8. "Learn" all the objects that you want to detect.

9. Check that your (mono) camera is publishing its output to a ros topic (sensor_msgs/Image message).

10. Considering that your camera publishes to /stereo/left/image_rect_color, launch the find_object_2d gui, keeping in mind that the image topic has to be remapped accordingly to your settings.

    rosrun find_object_2d find_object_2d image:=/stereo/left/image_rect_color
    

11. On the GUI, check that nextObjID under 'General' is set to 0.

12. Go to Edit->'Add object from scene...'

13. When you are satisfied with the position of the object in the scene, click 'Take picture'.

14. With the mouse, select the region of the object you want to detect (it's better that it contains a large number of keypoints to have a robust detection afterwards), click next and then end.

15. It is recommended to follow steps 7.iv and 7.vi for each object from several angles and distances.

16. Save the session by going to file->save session. Check that the name has to be *.bin. This saves all the objects learned in previous steps.

17. Once all the objects have been 'learned' (preferably from different angles and distances), create one text file (*.txt) per object and include in each one the ID number captured by the GUI for each angle. For example, if the captured images for object named 'phone' go from 0 to 2 and for 'fan' go from 3 to 5, then:

     phone.txt should look like this:
     0
     1
     2
     
     fan.txt should look like this:
     3
     4
     5
    

    Keep in mind that in each file, only one ID per line should be written and there should not be an empty line at the end.

18. All the .txt files have to be saved in the same folder.

19. Edit stereo.launch so it complies with your setting:

20. Check that the two image topics remaps are according to your left and right topics from your cameras.

21. Change the session_path parameters for both nodes to match the path to your session file (*.bin) saved in step 8.

22. Launch the stereo.launch file. It will start two instances of the find_object_2d (without the GUI)

    roslaunch find_object_2d stereo.launch 
    

23. Run the genom3 module

    objectdetection-ros -b
    

24. Run genomix

    genomixd &
    

25. Start TCL

    eltclsh
    

26. Load objecdetection module

    package require genomix
    ::genomix::connect
    genomix1 load objectdetection
    

27. Connect the necessary ports (Keep in mind that CameraL and CameraR have to comply with your cameras's setting)

    ::objectdetection::connect_port RightCameraParameters /stereo/right/camera_info
    ::objectdetection::connect_port CameraL /stereo/left/image_rect_color
    ::objectdetection::connect_port inObjectsL find_objects_2d_Left/objectsStamped
    ::objectdetection::connect_port CameraR /stereo/right/image_rect_color
    ::objectdetection::connect_port inObjectsR find_objects_2d_Right/objectsStamped
    

28. Run the detection activity

    ::objectdetection::Start {objectPath /path_to/objects/textfiles/}