BackGround/tutorials


[1] Ricker, Kim M. "Georeferencing: The Geographic Associations of Information." Portal : Libraries and the Academy 7.3 (2007): 387-8. ProQuest. Web. 4 Apr. 2013.

[2] Titova, O. A., and A. V. Chernov. "Method for the Automatic Georeferencing and Calibration of Cartographic Images." Pattern Recognition and Image Analysis 19.1 (2009): 193-6. ProQuest. Web. 4 Apr. 2013.

[3] Lin, Henry. “Clustering 15-381 Artificial Intelligence.”  Web. 9 Apr. 2013.


http://www.mathworks.com/discovery/MATLAB-opencv.html - for computer vision.


Below is the Design Proposal:




ENGR 103 - Spring 2013
Freshman Engineering Design Lab

“Satellite Image Processing”

Project Design Proposal

Date Submitted: April 10, 2013


Submitted to:
Anu Pradhan, arp69@drexel.edu
James Warcester, james.worcester2@gmail.com
Group Members:
Louis W. Rogowski, lwr24@drexel.edu

Dongen Zhou, dz78@drexel.edu

Eddie Tang, et354@drexel.edu





Abstract:
The purpose of this project is to use satellite image processing techniques to analyze civilian infrastructure after the events of hurricane Sandy. Analyzing the displacement of power lines is the primary focus of this project. The goal of this project is to write a program that will display the power lines located within the image so that any discrepancies can be located (missing or distorted power lines). The major tasks associated with this project include: georeferencing, filter application, k-means clustering, and the creation of a method to systematically perform the last three steps.  The technical challenges include finding the best algorithms to perform each step, understanding those algorithms, and writing an efficient code to perform the entire process. The final deliverables of this project will be the overall program (or a series of smaller programs) and the resulting images produced.   

1           Introduction


The devastation caused by hurricane Sandy completely crippled public infrastructure along the east coast. During the aftermath, a Cessna 172 manned aircraft flew over forty miles of the New Jersey coastline [1]. The photographs taken by the aircraft, multispectral aerial images (that include visible and long wave infrared and ultraviolet wavelengths), can be used to analyze the damage caused by Sandy to civilian infrastructure. The purpose of this project is to develop a program, utilizing image processing techniques, which can analyze these photographs and determine the extent of the damage caused to nearby power lines. This method will reveal any discrepancies in the power line infrastructure which will allow for quicker repairs.  

There are several learning objectives that will be achieved at the conclusion of this project. At the completion of this project the group members will have a better understanding of MATLAB, Arc ESRI, along with the different algorithms associated with image filtering and k-means clustering. This project will also strengthen communication between group members and provide more experience in a team based environment. Finally this project will give group members experience in carrying out an engineering project from beginning to end.  
There are several major tasks that are necessary for the project’s success: georeferencing, edge detection, filtering, k-means clustering, and combining object classification and combining the last three steps into an automated program. There are several technical challenges that need to overcome during the course of this project. The desired outcome of this project is to have a fully functional program that is capable of systematically performing filtering, clustering, and edge detection. The program needs to maintain a high degree of accuracy during the course of each step.

2           Deliverables

By the end of this project a fully functional program will be created that is capable of simultaneously performing edge detection and object classification, with a georeferenced image. After using ArcMap ESRI to georeference the image it will be run through the completed program. A series of filters will be used to create signatures for the different objects contained within the image. Then, the program will use k-means clustering to separate out the different signatures and only display only the signatures associated with power lines. Afterwards, the program will run the image through an edge detection algorithm, targeting only the signatures representing power lines that were gathered through the k-means clustering. The end result of this program will be georeferenced again to associate it with the area under observation. Results of the programs performance will also be available at the end of the project in order to benchmark the programs performance. This process may be broken down into a series of smaller programs (each one performing a specific step) instead of one holistic program.

3           Technical Activities

3.1         Georeferencing

The first major task of this project involves georeferencing the photographs with their corresponding positions using Google Earth as the base. Georeferencing aerial images allows it to be viewed, queried, and analyzed with other geographic data [1].“In simple terms, georeferencing just means relating information to geographic location”[2]. Specifically in processing satellite images, georeferencing is used to locate objects in the terms of geographical projections, or coordinating systems. The basic procedure for georeferencing is: “1. Scanning a paper topographic plan (map board) and obtaining a digital raster image; 2. Referencing control points (using a coordinate grid) and transformation; 3. Vectorization”[3]. This process will be performed using ESRI ArcMap software. This georeferenced image will then be run through a series of filter algorithms. At the conclusion of the edge detection (discussed later) the final result will be georeferenced again to the observed area.

3.2         Edge Detection:

MatLab is equipped with several functions that can be used to produce edge detection. The edges within an image are determined by locating points where there is a sharp gradient change between individual pixels. These gradients can be found using several different approximation methods such as the canny, sobel, and prewitt methods. Each of these methods finds edges that comply to certain threshold values; by manipulating these values it should be possible to isolate the edges of power lines.  The canny method produces the most detailed edge detection since “The method uses two thresholds, to detect strong and weak edges, and includes the weak edges in the output only if they are connected to strong edges. This method is therefore less likely than the others to be "fooled" by noise, and more likely to detect true weak edges” [4]. This method’s appears to be ideal for the project, however other methods may be chosen as the project progresses. Gabor filters may take the place of the above methods since they can be used for edge detection and assigning signatures to objects within the image.

3.3         Filters

Gabor filter are used in many applications involving texture segmentation, target detection, fractal dimension management, document analysis, edge detection, retina identification, image coding and image representation [5]. The strategy with these filters is to assign separate signatures for power lines and to the other objects within the photographs. These signatures should be unique to each pixel. These signatures will then be organized using a k-means clustering algorithm (discussed next). These filters will be accessed by using OpenCV in conjunction with MATLAB. This section will require a lot more research compared to the other sections since the mathematics behind these filters are complex. Finding the right filter(s) for this project will greatly affect the accuracy of the final product.

3.4         K-Means

Creating a correct K-means algorithm is one of the most important steps in this project. K-means is an algorithm where the n observations are being partitioned into K clusters, in the process of mining data, to give the closest mean value [6]. Similar observations will be placed into clusters that share the similar values linked to the observations. When the clusters are created, the center of the cluster is then moved to the mean of the data set. The objects within the cluster are then re-clustered to make sure everything fits in the data set. If the data set changes, the center has to be relocated for the data to be more accurate. The process is repeated until no change is required. Doing this with an appropriate cluster value should put all of the objects representing power lines into a single cluster. Using this information, edge detection can be used to outline all of the power lines present on the image. After this image is georeferenced with the area being observed, any damage to local power lines should be noticeable.  

4           Project Timeline
Week
Task
1
2
3
4
5
6
7
8
9
10
Literature study
x
x
x




x
x

Georeferencing

x
x







Edge Detection


x
x
x





Filtering





x
x
x


K-means






x
x
x

Final report preparation







x
x
x
Table 1: Time Schedule

5           Facilities and Resources

The resources required for the completion of this project are software packages and faculty assistance. The computer software that will be used for georeferencing is known as ArcMap ESRI, this program is not currently possessed by the group and will need to be downloaded. The majority of the programming will need to be done in MATLAB; this program is free for students at Drexel University. Libraries, such as OpenCV will be needed to access filters, edge detection, and other computer vision algorithms that will help classify objects within an image. OpenCV is free to download and interfaces with MATLAB to perform several instances of computer vision. The guidance from lab professors Pradhan and Warcester will keep the project on course and provide additional software and algorithm recommendations. Engineering labs and private study areas in the library will be used for workspace over the course of this project.

6           Expertise

            The majority of members have had previous programming experience, but only one has had significant experience in programming MATLAB. All three members bring something unique to the group and will be instrumental in the projects completion. Louis W. Rogowski has limited experience with MATLAB but has programmed in C++ and Maple in previous classes.
Eddie Tang is currently taking a MATLAB course and knows how to use a lot of the functions in MATLAB. Eddie has experience in C++, Java, and HTML, knowing such languages should help if any problems occur while using MATLAB. Dongen Zhou is also taking a MATLAB course related to the mathematical solution of differential equations. Dongen also has some experience with Maple in the previous CS class.



7           Budget


Category
Price
MATLAB (Software)
$0
ArcMap ESRI
$0
OpenCV
$0

7.1         MatLab

This is where the majority of the programming will occur. Matlab will be used to test the algorithms and run the completed code.

7.2         ArcMap ESRI

This program will be used to georeference the images taken by the Cessna 172 aircraft.

7.3         Open CV

This will provide MatLab access to computer vision functions and algorithms. Additional software packets may be necessary as the project progresses.

8           References

[1] Pradham, Anu & Warcester, James. “ENGR 103 064: Satellite Image Processing”. Web-Slides. 10 Apr. 2013.

[2] Ricker, Kim M. "Georeferencing: The Geographic Associations of Information." Portal : Libraries and the Academy 7.3 (2007): 387-8. ProQuest. Web. 4 Apr. 2013.

[3] Titova, O. A., and A. V. Chernov. "Method for the Automatic Georeferencing and Calibration of Cartographic Images." Pattern Recognition and Image Analysis 19.1 (2009): 193-6. ProQuest. Web. 4 Apr. 2013. http://search.proquest.com/docview/198144400/fulltextPDF/13D5671522071089A70/1?accountid=10559

[4] “MatLab User Manual”. Web. April 10, 2013

[5] Domke, Justin & Prasad, V. Shiv. “Gabor Filter Visualization”. Web. 10 Apr. 2013. http://www.cs.umd.edu/class/spring2005/cmsc838s/assignment-projects/gabor-filter-visualization/report.pdf

[6] Lin, Henry. “Clustering 15-381 Artificial Intelligence.”  Web. 9 Apr. 2013.

 

No comments:

Post a Comment

Subscribe to: Posts (Atom)