透過您的圖書館登入
IP:18.189.145.20
  • 學位論文

土石流影像前處理與機械視覺判釋之整合研究

Integrated Researches on Preprocessing and Interpreting for the Images of Debris Flow

指導教授 : 陳彥璋 張守陽

摘要


本研究藉由「機械視覺」(Machine vision)原理,針對影像前處理與土石流影像判識方法進行整合研究,希冀可提供土石流監測之參考。 經研究發現,若影像經「亮度等值化」,可降低外在環境(如霧、光源不足等)影響;「雜訊濾除處理」可有效降低胡椒鹽雜訊及高斯雜訊。「土石流波前判識」使用前處理影像,可有效的將土石流波前擾動強度放大,較易正確判識出土石流波前;「多重濾波判識」經影像前處理後,可降低因環境帶來的不當訊息,結果發現,其判識率由56.36%大幅提升至95.27%;「溪床特定物殘留率」判識時,易將雜訊視為特定物,濾除雜訊可有效降低誤判之機率。 由於現地CCD佈設不易垂直於溪床土石,大多僅能傾斜拍攝,因此本研究針對傾斜拍攝土石影像進行粒徑分佈之校正與分析,透過室內實驗加以驗證,首先進行「拍攝角度校正」,模擬CCD垂直拍攝之土石分佈影像,再分別以「連線遮蔽補遺」與「圓形遮蔽補遺」補遺礫石遮蔽部分,經拍攝角度校正與補遺修正後,中值粒徑與幾何標準偏差之誤差有所改善。本研究亦進行土石運移軌跡追蹤,首先進行影像分割,比較前後影像面積與位置之關係進行追蹤土石軌跡,進而推測土石運移速度,誤差約為3.63%。

並列摘要


The study based on machine vision techniques is to integrate preprocessing and detection for the images of debris flow. The histogram equalization technique can improve the contrast of the images degraded by environmental factors like fog, rain, and so on. The noise removal technique can effective remove the pepper-and-salt noise and Gaussian noise. The wave-front detection technique can amplify the variations of the wave-front of the debris flow. The debris flow becomes easy to discriminate. Based on the clean images processing, the detection ratio of the multiple filtering technique can increases from 56.36% to 95.27%. The noise removal technique can effective remove the pepper-and-salt noise and Gaussian noise. The clean images will reduce the false alarm rate for detection of remaining ratio technique. In order to the CCD is not likely installed perpendicularly to the river bed, it shoots at an angle for the most part. This study will correct and analyze the grain size distribution using the image shooting at an angle and prove it by indoor tests. At first, we adjust the shooting angle, simulate the image taken vertically, and add the part covered from the stones utilizing “connect shielded addendum” and “circular shielded addendum.” Both the median particle size and geometric standard deviation error reduce after adjusting the angle and correcting the addendum. This study also tracks the debris migration; we fulfill the image segmentation and track the debris migration comparing the relationships between comparative imaging area and location to speculate about the grain speed, the error approximately is 3.63%

參考文獻


[38] 劉榮斌,機械視覺與攝影測量判讀土石流特性參數之初步研究,碩士論文,國立台北科技大學環境規劃與管理研究所,台北,2006。
[8] 尹孝元、黃清哲、連惠邦、李秉乾、周天穎、王晉倫,「自動化土石流觀測系統之發展及應用」,中華水土保持學報,第三十七卷,第二期,2006,第91-110頁。
[24] 張守陽、林志平、游勝任,「影像處理應用於土砂災害判識之研究」,台南:電子計算計於土木水利工程應用研討會論文集,2005,第13-18頁。
[16] 吳榮峰,大尺度質點影像量測法之應用,碩士論文,國立成功大學水利及海洋工程研究所,台南,2003。
[9] 李明靜,河川表面流速與流量非接觸式量測方法之發展及應用,博士論文,國立成功大學水利及海洋工程研究所,台南,2003。

延伸閱讀