A 2D Shape Recognition Package for Applications in Weapon Detection
Advisor Information
Renat Sabirianov
Location
Dr. C.C. and Mabel L. Criss Library
Presentation Type
Poster
Start Date
7-3-2014 1:00 PM
End Date
7-3-2014 4:00 PM
Abstract
Modern systems for public safety (such as x-ray and full body scanners) would be revolutionized by a computer’s ability to automatically recognize the shapes of weapons in images. Thus, given an image of a detected object, we intend to compare its shape to a library of weapons in order to determine whether the detected object is indeed hazardous. We have developed a software package which innovatively recognizes the presence of a weapon in an image. We begin by isolating the imaged detected object from the background. We use a basic image processing subroutine to filter the image pixels into two color categories (black if inside the object, white if outside). We then extract the boundary (or contour) pixels for later analysis. Next, we smooth the resulting contour using a rapid iterative process (called a running average). To compare the detected object’s shape to that of a weapon in the library, we find the best overlap of the two objects using a correlation product, aided by a Fast Fourier Transform to increase the speed of computation. We further our comparison analysis by relating various shape equations, outputting a set of parameters indicating differences and similarities in shape. Such parameters include differences in overall area, perimeter, or local curvature of the contour. Finally, library objects are grouped into sub-libraries based on type and shape similarity, and a final decision is made as to indicate whether the detected object is statistically similar enough to be considered a weapon. We report test results of weapon detection for various imaged objects.
A 2D Shape Recognition Package for Applications in Weapon Detection
Dr. C.C. and Mabel L. Criss Library
Modern systems for public safety (such as x-ray and full body scanners) would be revolutionized by a computer’s ability to automatically recognize the shapes of weapons in images. Thus, given an image of a detected object, we intend to compare its shape to a library of weapons in order to determine whether the detected object is indeed hazardous. We have developed a software package which innovatively recognizes the presence of a weapon in an image. We begin by isolating the imaged detected object from the background. We use a basic image processing subroutine to filter the image pixels into two color categories (black if inside the object, white if outside). We then extract the boundary (or contour) pixels for later analysis. Next, we smooth the resulting contour using a rapid iterative process (called a running average). To compare the detected object’s shape to that of a weapon in the library, we find the best overlap of the two objects using a correlation product, aided by a Fast Fourier Transform to increase the speed of computation. We further our comparison analysis by relating various shape equations, outputting a set of parameters indicating differences and similarities in shape. Such parameters include differences in overall area, perimeter, or local curvature of the contour. Finally, library objects are grouped into sub-libraries based on type and shape similarity, and a final decision is made as to indicate whether the detected object is statistically similar enough to be considered a weapon. We report test results of weapon detection for various imaged objects.
Additional Information (Optional)
Winner of Meritorious Undergraduate Poster Presentation