應用影像辨識於自由水面偵測

當土石流進入畫面中會導致灰階值產生劇烈變動，目前已能透過計算灰階值平均改變速率成功預警土石流，因此本研究嘗試利用自定義的雜訊標準與實時影像之灰階值作比較，進而去除背景與環境雜訊，以偵測土石流的前鋒或表面，並計算流深，發揮監測作用。首先，在土石流預警前計算內之灰階值平均值與標準差作為背景值範圍，同時計算內每兩幀相減後每個像素的灰階值變動量標準，當土石流發生時，將原始畫面經前處理後的灰階值與背景值範圍比較，並將即時畫面前後兩幀相減得到每個像素之灰階值改變量，保留灰階值改變量大於灰階值變動量標準的像素，再透過中值濾波器消除剩餘離散分布之雜訊或較小型的移動物體，最後萃取水面以計算土石流流深。本文中分別使用室內實驗影片進行方法驗證與現地影像進行案例分析，在方法驗證中，依據捕捉保麗龍球前緣位置推算其移動速度，並與實際量測水流速度作比較；案例分析時則藉由偵測土石流在特定斷面之像素位置與人眼判斷結果進行誤差計算。根據結果顯示，在室內實驗影片驗證中，程式分析移動物位置時根據數量不同，其絕對誤差自0至3個像素，而由位置推算保麗龍球運動速度與實際流速之相對誤差百分比從最少可達1%；在實際案例分析中平均水面位置的絕對誤差介於1至14.6個像素，透過事件前計算背景灰階值及標準差作比較能夠成功去除影像雜訊，惟受環境亮度限制，在夜晚時無法進行分析，以及須進一步設立排除大型移動物造成偵測錯誤的措施。

關鍵字

土石流；影像處理；灰階值；影像雜訊；即時監測

並列摘要

In recent years, there have been many debris flow monitoring instruments set up in mountainous area in Taiwan by Soil and Water Conservation Bureau, such as camera, geophone, wire-sensor, etc. Among them, the camera immediately returns real-time images, which providing subsequent image processing for hazard early-warning. When debris flow gets into the image, the grey level will change drastically. At present, it has been able to successfully detect the occurrence of debris flow by calculating the change rate of the average grey level. Therefore, this study aims to use a custom noise standard to compare with the grey level of real-time images, and then remove the background and environmental noise, so as to extract the surface of debris flow, and eventually calculate the information such as flow depth. Initially, the average value and standard deviation of grey level within are calculated as background value range before the warning occurs. In the meantime, calculating the variation of each pixel between two frames as standard of grayscale variation. When debris flow happens, subtract the two successive frames of real-time images to obtain the variation of the gray level value of each pixel. Subsequently, make two comparisons to keep pixels that representing debris flow, one is checking if grayscale value of the original frame is out of the background range, and the other is if the variation of the grayscale value is greater than the standard of grayscale variation or not. Afterward, a few discrete distributed noises or small moving objects are removed through median filter, and the depth of flow could be calculated after edges selected in the end. In this research, the laboratory experiment videos are used for verifying the method, while the actual debris flow images are used for case analysis. After capturing the forward position of styrofoam ball by the method, estimating its moving velocity and comparing with the velocity of the flow in the channel. In case analysis, comparing the pixel position detected by means of program with judgement result of human eyes. According to the result, the absolute error of between program result and eye recognition ranges from 0 to 3 pixels, and the minimum relative error percentage between the velocity of the flow and which is calculated from the position of the styrofoam ball reaches 1%. In addition, the absolute error of the average water surface position is between 1 and 14.6 pixels in the actual case analysis. Overall, noise can be successfully removed by comparing with the background grayscale value and standard deviation which are calculated before the event. However, due to the limitation of ambient brightness, the analysis cannot be performed at night, and further measures must be established to eliminate detection errors caused by large moving objects.

並列關鍵字

debris flow ； image processing ； grayscale ； image noises ； real-time monitoring

參考文獻

Adrian, R. J. (1991). "Particle-imaging techniques for experimental fluid mechanics." Annual review of fluid mechanics 23(1): 261-304.

Google Scholar

Fujita, I., et al. (1998). "Large-scale particle image velocimetry for flow analysis in hydraulic engineering applications." Journal of Hydraulic Research 36(3): 397-414.

Google Scholar

Creutin, J., et al. (2003). "River gauging using PIV techniques: a proof of concept experiment on the Iowa River." Journal of hydrology 277(3-4): 182-194.

Google Scholar

Theule, J. I., et al. (2018). Exploiting LSPIV to assess debris-flow velocities in the field. Natural Hazards and Earth System Sciences. 18: 1-13.

Google Scholar

Pham, C. C., et al. (2011). Adaptive guided image filtering for sharpness enhancement and noise reduction. Pacific-Rim Symposium on Image and Video Technology, Springer.

Google Scholar

主題瀏覽