jupidity / svm_model_generation

Gazebo ROS model to train RGBD SVM model

SVM Model Generation PACKAGE

---

This is a Gazebo and ROS package for generating a model.sav of color and normals for various objects contained in the /models directory fir use in object recognition. The package was modified from the work of Brandon Kinnman and Harsh Pandya from Electric Movement and Udacity RoboND.

Depends

---

• Gazebo 1.7
• ROS Kinetic
• sklearn
• numpy
• pcl 1.8

Use

---

This assumes you have a ~/catkin_ws/src directory structure.

clone the repo

cd ~/catkin_ws/src/
git clone https://github.com/jupidity/svm_model_generation.git

make in your workspace

cd ~/catkin_ws
catkin_make

if you wish to generate features from a particular model, it must be added to the /models directory

otherwise, launch the training simulation

roslaunch svm_model_generation training.launch

a stick with an RGBD camera should appear in Gazebo sim. Next lauch the capture features node.

rosrun svm_model_generation capture_features.py

The model will be generated from 40 captures of the object in a rondom orientation with 256 histogram bins of HSV color space and geometric normals.