point-cloud-recognition

This project captures features of objects and then train a classifier to correctly identify the objects from a point cloud file.
This is my solution for the last perception exercise from Udacity's RoboND.
This builds upon my solution for the second perception exercise where I use the DBSCAN algorithm to cluster points pertaining to the same object.

Important Files

You need an Ubuntu 16.04.2 with ROS full-desktop-version which includes RViz and Gazebo
You must clone the repository, go inside the directory and install the dependencies:

$ rosdep install --from-paths src --ignore-src --rosdistro=kinetic -y
$ catkin_make

export GAZEBO_MODEL_PATH=~/catkin_ws/src/sensor_stick/models
source ~/catkin_ws/devel/setup.bash

pip install sklearn scipy

First you must capture features.
Run this on one terminal. You should see an empty scene in Gazebo with only the sensor stick robot.

$ roslaunch sensor_stick training.launch

$ rosrun sensor_stick capture_features.py

This script spawns each object in random orientations (default 20 orientations per object) and computes features based on the point clouds resulting from each of the random orientations. The features will now be captured and you can watch the objects being spawned in Gazebo. It should take 5-10 seconds for each random orientations (depending on your machine's resources) so with 7 objects total it takes awhile to complete. When it finishes running you should have a training_set.sav file.
You can inspect the script here where you will see that NUMBER_OF_SPAWNS = 20
- https://github.com/mithi/point-cloud-recognition/blob/master/src/sensor_stick/scripts/capture_features.py
The features captured are the normalized histogram of the colors of the three channels in HSV format as well as the normalized histogram of the normal vector at each point which capture the shape
You can inspect how the features being retrieved from the point cloud here.
https://github.com/mithi/point-cloud-recognition/blob/master/src/sensor_stick/src/sensor_stick/features.py
After this run the following:

$ rosrun sensor_stick train_svm.py

$ roslaunch sensor_stick robot_spawn.launch

python object_recognition.py