SPOC-Lite

SPOC-Lite (Soil Property and Object Classification Lite) is a light-weight terrain classifier developed in FY16/17 Topical R&TD Next-Gen AutoNav.

Prerequisite

• Ubuntu 14.04
• ROS Indigo
• catkin_tools (aka catkin build)

Installation

Get the code

Checkout this repository and claraty_comm to your ROS workspace.

If you use wstool, this can be automated as

mkdir ~/spoc_ws && cd ~/spoc_ws
wstool init src http://claratyhub.jpl.nasa.gov/maars/spoc_lite/raw/master/spoc_lite.rosinstall

You should see spoc_lite and claraty_msgs under src directory.

Build

catkin build

Usage

Launch terrain classifier pipeline

roslaunch spoc_lite launch.launch camera:=my_camera image:=image_rect

Run demo program

roslaunch spoc_lite demo_classifier.launch
roslaunch spoc_lite demo_mapping.launch

Create a terrain classification model from training data

TBD

Camara parameter settings

The model used in the demonstrations was made from images taken at 2017 March, which was very sunny day. Depending on the weather condition you run spoc_lite, you may need to tune one camera parameter "exposure". Below is how-to.

1. After you launch "spoc_lite", launch rqt.
2. Go to configuration -> "XXX"
3. Set the camera parameters as below.
   • auto_brightness: manual(3)
     • brightness: 0.0 (this is default value)
   • auto_focus" None(5)
     • focus: 0 (this is default value)
   • auto_gain: manual(3)
     • gain: 178 (this is default value)
   • auto_gamma: Off(0)
     • gamma: 10.0 (this is default value)
   • auto_hue: off(0)
     • hue: 2048 (this is default value)
   • auto_iris:None(5)
     • iris:8.0 (this is default value)
   • autop_pan:Off(0)
     • pan:4 (this is default value)
   • auto_saturation: Manual(3)
     • saturation: 1278 (this is default value)
   • auto_sharpness: Manual(3)
     • sharpness: 1532 (this is default value)
   • auto_shutter: Auto(2)
   • auto_white_balance: Manual(3)
     • white_balance_B: 794
     • white_balance_R: 487
4. Finally set the parameter "exposure"
   • auto_exposure: manual(3)
     • exposure: 300-600 (depending on the weather condition, tune this parameter)

Package documentation

sand_classifier

Main terrain classifier node.

ROS API

ortho_projector

Orthogonal image projection onto local groud plane.

ROS API

Contact

If you have any questions, contact the authors:

• Yumi Iwashita ([email protected])
• Kyohei Otsu ([email protected])