Recognizing objects

no-virtual Cannot be tested on a simulated robot, unless in a virtual world.



Teaching NAO to recognize objects

Using the Video monitor panel, NAO can learn images, objects and rooms to recognize.

Step Action
Make sure you are connected to a real robot or to a simulated robot evolving in a virtual world.

Choose View > Video monitor.

The Video monitor panel appears and displays what the active camera is seeing.

Click the Learn learn-button button.

A 4 seconds countdown starts, it gives you time to place correctly the object you want to learn.

../../../_images/capturing-vision.png

At 0, the image switch to QVGA resolution and is captured.

Click to draw, segment by segment, the contour of the object you want to learn.

../../../_images/contour-vision.png

Note

For a room, take the contour of the full image.

As soon as you click again the first point to close the shape, a pop-up opens.

Enter the information about the object, book or room:

../../../_images/object-tag.png

Click OK.

A feed-back message confirms if the learning is successful or not (due to bad quality of the input).

If you want to learn other objects:

  • Click the Play play-button button and
  • return to step 2.

When you are done with learning objects, you can:

  • Click the Send current vision recognition database to NAO send-database-button button, in order to use immediately the database on the robot, or
  • Click the Export Vision Recognition Database export-database-button button, in order to save the database on your computer for later use.

Note

The database is not stored with the behavior but in a specific directory. To create several databases with different object lists or share it on other computers, you can clear, export and import the database using the dedicated buttons in the video monitor. For further details, see Video monitor panel functionalities.

Recognizing objects

Once a vision recognition database is created and launched on the robot, NAO can recognize the objects defined in the database.

Step Action
Make sure you are connected to a real robot or to a simulated robot evolving in a virtual world.

Make sure you have launched a vision recognition database on the robot.

If not:

  • Create a vision recognition database -or import a previously stored one- and
  • Click the Send current vision recognition database to NAO send-database-button button.
Create a behavior using the Vision Reco box.
Launch it so NAO can detect the newly learned objects.

You can also create a Python script: ALVisionRecognition Tutorial.