以影像偵測土石流前鋒

Abstract

在土石流監測站中，攝影機為主要的監測儀器之一，為了加值利用影像資料，本研究藉由土石流發生時，巨礫受重力作用集中於波前的特性，依據人眼視覺以灰階值劇烈變化作為判釋基礎，應用於影像偵測土石流，並期望此方法能提供未來發展土石流即時偵測系統之參考。 首先，於影像中河道區域繪製一個ROI(Region of interest)，計算ROI面積中之平均灰階值及隨時間之改變速率，並以改變速率作為事件之偵測指標；再依照土石流於空間、時間中的特性，建立土石流前鋒之偵測條件，土石流發生時的平均灰階值變化相對於環境亮度改變來的劇烈且快速，故以任一偵測時間點前一段時間的環境亮度變化速率最大值 作為基準值，並以基準值的倍率 作為偵測時間點的門檻值，另外，因為土石流前鋒通過偵測斷面後會延續一段時間，因此本研究將偵測時間點之斜率連續大於門檻值一段時間，作為偵測土石流事件發生與否的判斷條件。由於此方法以灰階值改變速率作為偵測因子，環境光源強度為影響灰階值之因子，故設計一室內水槽實驗，以四種不同環境亮度討論此方法對於光線的敏感度；又以室內土石流實驗，討論資料平均延時、門檻值倍率、影片幀率對偵測結果的影響；最後以兩個現地案例，討論ROI的選取與否對於事件偵測的影響，並且是否可以應用於影片幀率較少的情況。 根據本研究結果顯示，總灰階值法應用於兩個室內實驗中，資料平均延時為1或2秒對於事件偵測的精確度無顯著影響，而門檻值倍率分別以 、5的條件下可得最佳偵測時間，並且環境中只需要有微弱的光線，即可應用本方法於事件偵測；於現地影片的分析中，僅 的條件下可偵測到土石流事件，又當影像中存在大面積非土石流流動區域時，ROI的選取有其重要性，另外，在長時間的土石流監測，即使監測站回傳的數據量較少，本法仍有機會應用於偵測土石流事件。

Cameras are one of the major instruments used in the debris flow monitoring stations. By means of the image data from those cameras, this study is focused on detecting the change in grey level because of boulders concentrated on the wave front due to gravity when the debris flow occurs. These sharp changes in grey level which are also detected by human eyes are used as the basis for interpretation and applied to image data. This is expected to provide an outline to develop a real-time detection system for debris flow in the future. Initially, a region of interest (ROI) is selected in the image. Then, the average grey level of selected ROI and its change with respect to time was calculated, which was used as an indicator of event detection. Subsequently, according to the spatial and temporal characteristics of debris flow, the detection condition of debris flow front was established. The change in average grey level of debris flow is rapid and intense as compared to the environmental brightness. Therefore, the maximum ambient brightness of certain time interval before detection was taken as the reference value, which when multiplied by a constant (Alpha) gives us a threshold value for detection. Since the front of debris flow will continue for a period of time after initially passing through the ROI, the slope of the average grey level after detection should be continuously larger than the threshold for a period of time. This can be used to differentiate the actual event from sudden environmental changes. As this method depends upon the change of grey level, the intensity of ambient light source is a major influencing factor. Therefore, an indoor experiment was designed with four different environmental lighting conditions to analysis the effect of ambient lighting on the grey level during an event. Moreover, we discuss the effect of period of time average, multiplier for threshold (Alpha), and frame rate of the camera on the detection of the event. Likewise, using the actual data from the field, the impact of ROI selection and application for the low frame rate videos are also discussed. Based on the results of indoor experiment, small change in period of time average by 1 to 2 seconds had no significant effect on the accuracy of event detection and the multiplier for threshold (Alpha) of 2 and 5 presented the best results for two different indoor experiments respectively. The method was found to be successful for event detection even during very low environmental light conditions. However, the event detection was only possible from field data when the value of = 2. When the field of view of the camera included large static area, the choice of ROI within the path of debris flow was found to be very important. Furthermore, this method was also found to be successful even for low frame rate videos, although longer duration of data was required for effective detection of debris flow events.