本研究藉非接觸式之「機械視覺」與「攝影測量」理論,針對不同情境之土石影像分析土石之波高、表面運移速度及粒徑分佈等土石特性。於實驗室與土石流潛勢溪流現場分別架設雙攝影機進行影像擷取,並自行撰寫程式估算上述土石之特性。為說明程式之可行性及探討量測準確度,研究中先配合室內實驗與前人研究成果進行程式驗證與修改,進行土石特性之初步研究,希冀得以提供日後土石流特性監測及防災應變等參考。 機械視覺方面,以webcam對堤岸標示利於機械視覺判斷之不同顏色區塊,以ROI灰階值進行「波高」高程判讀;另外以雙線性內插推求不同土石顆粒三維座標,依時間距推求「表面運移速度」;粒徑分析方面,本文建立了一傾斜拍攝影像之校正理論,並透過室內實驗加以驗證,以模擬攝影機垂直拍攝土石表層之畫面。透過攝影機俯視拍攝,對影像進行傾斜與比例校正,分別藉由「球形模擬投影法」與「數位影像處理」理論分析各顆粒之代表粒徑,繼而求得土石表層粒徑分佈。攝影測量則以雙攝影機分析土石之三維空間座標變化,繼而估算土石「波高」及「表面運移速度」。 經由室內初步實驗結果,webcam堤岸波高判讀以移動平均ROI 灰階值判讀均可成功判讀;攝影測量在距離目標物4公尺時,其X、Y、Z方向的平均誤差約為0.56、0.41及0.66公分,若以立體物體進行測量,其X、Y、Z方向的平均誤差約為0.22、0.45及0.68公分;粒徑分佈實驗在標準乒乓球條件下,距離目標1.5公尺,以「球形模擬投影法」和「數位影像處理」理論進行計算,攝影機傾斜拍攝粒徑之最大誤差約為5.75% ~ 3.67%;在透過室內實驗後,希冀本實驗之初步結果可提供未來土石流特性之監測參考。
This study is to construct the theory of “Machine Vision” and “Photogrammetry” applied on debris. Under various scenarios, the characteristics of debris flow were analyzed. The characteristics studied herein included that the elevation of wave front, the floating velocity of specific object and the size distribution of debris flow. Two CCD Cameras installed at the laboratory and the debris flow potential streams site were to capture image to estimate size distribution of debris by computer programming. To examine the feasibility of the developed model, indoor experiment and previous empirical results were to verify the theory of “Machine Vision” and “Photogrammetry” constructed in this study. Hopefully the image differentiating of characteristics of the debris flow can be practically applied to the debris flow hazard mitigation in the further. Firstly, the CCD camera to capture image that the tide stave gray level to estimate the elevation of wave front. The image coordinate can transformed the three-dimensional space coordinate by bilinear interpolation. Then, the floating velocity of specific object in the image can be estimated accordingly. Indoor experiment Set up the slant image calibration theory to simulate vertical captured image, size distribute results were analyzed by the theory of “Globe Simulate Projection” and “Digital Image Processing”. As the three-dimensional space coordinate of wave front was obtained, the elevation of wave front can be therefore determined. The preliminary tests of the tide stave gray level can success to estimate by moving average. “Photogrammetry” show that the average error of estimated distance along the X, Y, Z direction is about 0.56, 0.41, 0.66 cm and the error of that is about 0.22, 0.45, 0.68 cm as the objects are three-dimensional respectively. By detecting a group of ping-pong balls, the error of the theory of “Globe Simulate Projection” and “Digital Image Processing” are around 5.75% ~ 3.67%.