首頁 > 網路資源 > 大同大學數位論文系統

Title page for etd-0221118-174530


URN etd-0221118-174530 Statistics This thesis had been viewed 30 times. Download 0 times.
Author Rong-En Hsu
Author's Email Address No Public.
Department Computer Science and Enginerring
Year 2017 Semester 1
Degree Master Type of Document Master's Thesis
Language zh-TW.Big5 Chinese Page Count 52
Title Visible and Infrared Images Fusion based on Dynamic Programming Matching using Contours of Moving Objects
Keyword
  • dynamic programming
  • image fusion
  • infrared image
  • Visible image
  • contour matching
  • contour matching
  • Visible image
  • infrared image
  • image fusion
  • dynamic programming
  • Abstract In recent years, infrared (IR) camera is often used to quickly measure the tempera-ture of passengers in public place. The captured image from an IR camera is different from that by a regular camera. Therefore, we need to match the objects (persons) in regu-lar RGB image with those in IR image. In the registration process, background subtrac¬tion is often used to get the moving objects in IR and visible images. The outline of the moving object is approximated by a sequence of line segments with angles in-between. Previous researches used length and angle of contour line segment for matching. Homog-raphy matrix is then used to transform the IR image for fusion with visible image.
    We proposed using dynamic programming to solve the contour matching problem between visible and IR image. Firstly, background subtraction is used to obtain the mov-ing objects in the visible image. Since the IR device is heat-sensitive and responds quick¬ly to changes in temperature, objects of higher temperature could be easily extracted by a simple threshold. HSV color model is superior to RGB model for getting those objects. Hue channel is used to obtain target objects even if background temperature changed. Dynamic programming is then used to calculate the difference of the feature points of contours. Next, we used the area and the difference of contours for matching.
    Finally, the images of IR objects are respectively transformed using perspective pro-jection and positioned to the most similar visible objects. Four test videos were captured from different place and the accuracy of the superimpo¬sition of IR and visible images is 80% in average which demonstrated the feasibility of proposed method.
    Advisor Committee
  • Chen-Chiung Hsieh - advisor
  • En-Hui Liang - co-chair
  • Mu-Chun Su - co-chair
  • Shang-Lin Hsieh - co-chair
  • Files indicate accessible at a year
    Date of Defense 2018-01-23 Date of Submission 2018-02-22


    Browse | Search All Available ETDs