13. Recognition Exercise

Recognition Exercise

In this exercise, you will continue building up your perception pipeline in ROS. Here you are provided with a very simple gazebo world, where you can extract color and shape features from the objects you segmented from your point cloud in Exercise-1 and Exercise-2, in order to train a classifier to detect them.

Your goal in this exercise is to train an SVM to recognize specific objects in the scene. You'll accomplish this by first extracting a training set of features and labels, then training your SVM classifier, and finally using the classifier to predict what objects are in your segmented point cloud.

If you already cloned the RoboND-Perception-Exercises repository, all you need to do is git pull again there to get the code for Exercise-3. Note: This exercise requires ROS, so you need to complete these steps in the VM provided by Udacity or on your own native Linux/ROS install.

Getting Setup

  • If you completed Exercises 1 and 2 you will already have a sensor_stick folder in your ~/catkin_ws/src directory. You should first make a copy of your Python script (segmentation.py) that you wrote for Exercise-2 and then replace your old sensor_stick folder with the new sensor_stick folder contained in the Exercise-3 directory from the repository.

  • If you do not already have a sensor_stick directory, first copy / move the sensor_stick folder to the ~/catkin_ws/src directory of your active ros workspace. From inside the Exercise-3 directory:

$ cp -r sensor_stick/ ~/catkin_ws/src/
  • Make sure you have all the dependencies resolved by using the rosdep install tool and running catkin_make: 
$ cd ~/catkin_ws
$ rosdep install --from-paths src --ignore-src --rosdistro=kinetic -y
$ catkin_make
  • If they aren't already in there, add the following lines to your .bashrc file 
$ export GAZEBO_MODEL_PATH=~/catkin_ws/src/sensor_stick/models
$ source ~/catkin_ws/devel/setup.bash

You're now all setup to start the exercise!