Title

Online Visual Water Differentiation Using Unmanned Aerial Vehicles

Poster Number

4

Lead Author Affiliation

Engineering Science

Lead Author Status

Masters Student

Second Author Affiliation

ECPE

Second Author Status

Faculty

Introduction

Scientists studying wetlands need detailed maps of the region to ensure model accuracy. These maps need to identify important features, as well as have regular updates to accurately reflect a constantly changing environment. Achieving detailed, current contours is hard without automation; scientists cannot manually re-map the wetlands after each season nor achieve the hoped-for details. Automated topographic map creation requires different sensors, algorithms, and processes. Determining which to use requires a map defining land and water locations, also requiring automation.

Purpose

We solve the problem of automatically generating a map of the environment by using aerial robots. These robots are capable of taking aerial images so that they have data that accurately represents the current state of the environment. They are also minimally invasive, only interacting with the environment in the locations where any sensors would be deployed. This reduces the risk of damaging the environment while deploying the sensor network.

Method

First we develop a system to identify land and water regions in aerial images. Our solution develops camera-based methods of clustering and classifying like locations. The approach allows for the unmanned aerial vehicle (UAV) to identify water and land locations while flying through the region. This allows immediate task allocation to other UAVs to install sensors to monitor the region. We develop the classifier through online training by k-means clustering a set of photos gathered by the UAV, labeling the clusters, training 12 classifiers, and identifying the best based on classification rate on the verification set as well as computational complexity. This classifier then forms a portion of the overall algorithm that takes a picture, clusters the picture, classifies the region, and then communicates the regions of interest to other UAVs. We also analyze the operational use of the algorithm to determine the limitations on flight speed, time between photos, and UAV height. We implement this algorithm on the UAV and perform a field experiment to verify its use over the Calaveras River in Stockton, CA.

Results

When the data was separated into 5 clusters, the LDA training algorithm performed the best, with a 91.33% success rate on the verification set of clusters that had been hand labeled. As the number of clusters increased however, LDA was replaced by the Random Tree training algorithm with 89.67% and 89.33% successful classification for 10 and 15 clusters respectively. The system successfully transformed the water regions into a set of polygons that could be stored as a line map.

Significance

The map generation system using unmanned aerial vehicles (UAV) provides an automated solution for mapping an area. This can help reduce the overhead required for projects that involve an ever changing environment, since the UAV can map the environment on a regular basis to ensure that the data used is representative of the current state of the environment. The generated maps can be provided to other UAVs that can perform tasks in the environment, such as deploying a sensor network.

Location

DUC Ballroom A&B

Format

Poster Presentation

Poster Session

Afternoon

This document is currently not available here.

Share

COinS
 
Apr 29th, 1:00 PM Apr 29th, 3:00 PM

Online Visual Water Differentiation Using Unmanned Aerial Vehicles

DUC Ballroom A&B

Scientists studying wetlands need detailed maps of the region to ensure model accuracy. These maps need to identify important features, as well as have regular updates to accurately reflect a constantly changing environment. Achieving detailed, current contours is hard without automation; scientists cannot manually re-map the wetlands after each season nor achieve the hoped-for details. Automated topographic map creation requires different sensors, algorithms, and processes. Determining which to use requires a map defining land and water locations, also requiring automation.