透過您的圖書館登入
IP:3.144.212.145
  • 學位論文

影像視覺之光譜照度分析與物件特徵辨識

Characteristics of Spectral Illumination and Object Recognition for Image Vision

指導教授 : 張文陽
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


本研究為影像視覺之光譜照度與紋理特徵辨識分析,主要探討影像在不同的光譜照度之特性與不同物件之幾何特徵檢測。影像的光譜照度分析為物件在均等照度下,對於光源照射的角度、強度、穿透率與光譜之調適性進行分析。物件特徵辨識主要針對物件幾何尺寸、真圓度、紋理特徵與影像接合等進行特徵檢測,其中分析的物件主要有扣件(墊片與螺紋)、硬幣與自行車零組件。墊片的檢測是藉由影像形態學量測物件的幾何尺寸、真圓度與瑕疵處,而在墊片尺寸量測中,外徑的量測誤差為±1%,而經影像處理的內徑之量測誤差也僅為±2.5%,其中墊片的瑕疵檢測是藉由XOR影像相減作為基礎,並成功的將墊片內徑之突起瑕疵物辨識出來。螺紋紋理的辨識為使用傅立葉頻譜進行頻域之轉換,再利用Sr(θ)與Sθ(r)的頻譜響應,即可順利區分良好螺紋與毀壞螺紋,其中螺距的檢測為針對M10×1.5、M8×1.25、M6×1.0與M4×0.7四種螺紋進行量測,在最後量測的誤差中M10×1.5與M8×1.25為±0.01mm,而M6×1.0與M4×0.7則是±0.03mm的誤差。硬幣的辨識首先以圓形裁剪將硬幣與背景影像進行分離,其次利用Sobel與影像相減判斷硬幣之正反面,最後使用平均對比度與平滑度識別出新舊硬幣。本研究另針對狹長型的自行車零組件,並以SIFT影像接合使其具備特徵完整性與高突顯性,且同時進行管長與管寬的量測,其量測誤差皆可控制在±1%內。

並列摘要


This study investigates the characteristics of spectral illumination and object recognition for image vision. In general, the image inspections for object recognition usually depend on different spectral illumination and geometric features. Therefore, the angle, intensity, transmittance and spectral analyses of light sources are analyzed for image inspections. The objects for image inspections have the fasteners (gasket and thread), coins and bicycle accessories. The feature inspections of the object contain the geometry size, roundness, texture feature and image stitching that are recognized using the image morphology. Results showed that the defect inspections of the gaskets based on XOR image subtraction are successfully recognized. The inspection errors of the outside and inside diameters for the gaskets are ±1% and ±2.5%, respectively. The frequency domain transformation of Fourier spectrum is used to identify the texture of the screw thread. Moreover, the spectrum response of Sr(θ) and Sθ(r) can recognize the destructive screw thread. The pitch error of the M10×1.5 and M8×1.25 is ±0.01mm and the error of the M6×1.0 and M4×0.7 is ±0.03mm.For the recognition of the coins, we use a circle crop to separate the image of coin and the background. Then, the head or tail of the coins is determined using the Sobel detection and image subtraction algorithm. Finally, the new or old coin is recognized using the average contrast and the smoothness. Additionally, the bicycle accessories with narrow type are investigated using scale-invariant feature transform (SIFT) algorithm. The dimensional error of the bicycle accessories is ±1% for the length and the width.

參考文獻


[14]呂志誠,鄭振宗,張文柳, "應用渦電流感測及統計圖樣分析法於硬幣識別技術",國立台北科技大學機電整合研究所碩士論文,2011
[30]林明宗,林聖泉,"新世紀梨重量量測裝置之研製",農林學報,2006年
[6]張晉瑋,李炳寅,"刀具幾何影像檢測系統之切削角量測開發",國立虎尾科技大學機械與電腦輔助工程系碩士論文,2012
[5]黃俊銘,"應用數位影像處理於焦炭粒度檢測",國立成功大學機械工程研究所碩士論文,2009
[3]許裕騄,孫永年,"棒鋼表面缺陷自動化檢測系統",國立成功大學碩士論文,2007年

被引用紀錄


李彥儀(2015)。基於田口法與FPGA平行運算辨識物件顏色之影像處理〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://doi.org/10.6827/NFU.2015.00030
楊孟哲(2015)。成型機主軸檢測系統建構〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://doi.org/10.6827/NFU.2015.00007
陳則豪(2014)。車管立把之FPGA即時影像自動化取放系統〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://doi.org/10.6827/NFU.2014.00054
楊承翰(2014)。基於FPGA之物件影像自動化辨識與檢測系統〔碩士論文,國立虎尾科技大學〕。華藝線上圖書館。https://doi.org/10.6827/NFU.2014.00021
Hung, C. H. (2017). 藉由共振柱試驗測定林口臺地不飽和紅土在小應變下的剪力模數及阻尼比 [master's thesis, National Taiwan University]. Airiti Library. https://doi.org/10.6342/NTU201703088

延伸閱讀