Image Analysis for Unmanned Vehicle

Project Leader: Kiwon Sohn

Goal: Image Analysis for Unmanned Ground Vehicle

Features: 1. Using Be Franklin Teams' Gaussian Approach

2. Implemented by Matlab

3. This algorithm is adopted by our small sized UGV with A Howard's approach

Abstract of this project:

One of important goals of Mobile Robot area is to make an Unmanned Ground Vehicle (UGV)

which can navigate safely and to arrive at certain goal point while moving automatically in

known and unknown environments. To accomplish that, the robot should have a capability of

analyzing terrain which is front of itself and determining whether terrain is traversal or not. In

this study, two different image segmentation techniques which The Ben Franklin team in UPenn

and Howard in JPL have developed and been using for their mobile robot in real environment

are studied. Both of they can distinguish ground area and obstacle region from various kinds of

terrain image and provide data which is critical to navigation of unmanned vehicle. Then,

experimental results using real outdoor road image data are provided by images and video clip below.

Experimental Result:

1) Test Image

2) Results of Top Left test image

Top Left result is using Single Gaussian approach of Ben Franklin Team.

Top Right result is using Mixture Gaussian and EM approach of Ben Franklin Team.

Bottom result is using approach of A Howard team.

3) Results of Top Right test image

4) Results of Bottom test image

5) Performance Comparison

6) Video