基於FPGA之物件影像自動化辨識與檢測系統

Automatic Recognition and Inspection System Based on FPGA Image Processing for Accessory

楊承翰(Cheng-Han Yang)

指導教授 ： 張文陽

國立虎尾科技大學/工程學院/機械與電腦輔助工程系碩士班/碩士(2014年)

https://doi.org/10.6827/NFU.2014.00021

摘要

本研究為基於FPGA之物件影像自動化辨識與檢測系統，即以影像自動化辨識零組件之位置及方向性的檢測，研究內容主要探討物件影像前置處理、幾何尺寸檢測、光源光譜特性與物件方向性角度辨識之FPGA演算法則。影像處理技術包括影像的灰階化、二值化、形態學處理、物件邊緣偵測與顏色辨識等。然而，影像邊緣偵測常因外在環境光源因造成物件邊緣顯示不完整，此外在因素無法經由調整閾值達到理想結果，必須調整環境之光源使邊緣偵測之邊緣有所改善，故本研究使用光譜儀量測物件周邊照度之強度與均勻性並適度做補光。在幾何尺寸量測實驗中，總共對車管立把、副把手兩種不同零組件尺寸進行量測，因車管立把中間部分為橢圓，所以共量測到兩種不同直徑，而對副把手只量測一種直徑，並使用機器視覺量測到的直徑分別為32.66mm、35.08mm與22.62mm，與實際尺寸比較下，其誤差分別為1.78%、0.52%與0.44%。在物件方向性辨識實驗中，以邊緣線段之斜率估算零組件方向性，對於方向性實驗中主要測得角度誤差與再現性，從實驗結果得知誤差值最高為1.79%，於方向性角度的再現性量測中，測得的再現率則為97.25%，另外，在幾何中心量測的誤差值為10.53%。

關鍵字

FPGA ； 影像處理 ； 自動化辨識 ； 邊緣偵測 ； 方向性檢測

並列摘要

This study is automatic recognition and inspection systems of objects based on FPGA image processing. The main researches are the FPGA algorithms of the image preprocess, morphology size measure, property of spectral illumination and angle recognition. The image processes include the gray scale, binary, morphology processing, edge detection and color recognition. However, the results showed that the image had broken edge for edge detection. The edge detection gets the bad results because of the illumination intensity and the light source uniform. For fixed the broken image, the light sources are the key point. We can adjust the light source to get the good light uniform for edge detection. In the morphology measurement, the experiment measured three type width sizes on two objects and the width sizes are 32.09mm, 34.90mm, 22.52mm by the vernier caliper. And measurements of the width sizes are 32.66mm, 35.08mm, 22.62mm by the machine vision. The inspection errors of the 32.09mm, 34.90mm and 22.52mm are 1.78%, 0.52% and 0.44%, respectively. In the measurement of angle, the accuracy and reproducibility was measured in the experiment. In the measurement of angle accuracy, the objects angle will be measured by image process before used the angle gauge to correct the angle. The results showed that the accuracy between 98.21%~99.83%. The angle reproducibility will repeat measure the same angle and get the angle value every time. The results showed that reproducibility are 94.17% and 97.25% with non-fill light and fill light.

並列關鍵字

FPGA ； Image Processing ； Automatic Recognition ； Edge Detection ； Inspection Systems

參考文獻

[6]林祺勛，李炳寅，"五軸工具磨床線上刀具幾何影像檢測系統開發"，國立虎尾科技大學碩士論文，2011年

[19]蔡治平，張文陽，"影像視覺之光譜照度分析與物件特徵辨識"，國立虎尾科技大學碩士論文，2013年

[1]王柔雯，劉伯祥， "直接法與影像量測法之比較研究：以量測外耳尺寸為例"，聖約翰科技大學碩士論文，2007年

[8]X.D. Bai, Z.G. Cao, Y. Wang, Z.H. Yu, X.F. Zhang, C.N. Li,"Crop segmentation from images by morphology modeling in the CIE L*a*bcolor space", Computers and Electronics in Agriculture 99 (2013) pp.21-34

[9]LiJun Chen, KeQi Wang, YongHua Xie, JinCong Wang,"The segmentation of timber defects based on color and themathematical morphology", Optik 125 (2014) pp.965-967