A Feasibility Study of Machine Vision Applied on Detecting the Sediment Hazards

• Main Authors: Jian-Shing,Wu, 吳建興
• Other Authors: Shou-young,Chang
• Format: Others
• Language: zh-TW
• Published: 2004
• Online Access: http://ndltd.ncl.edu.tw/handle/16348207224205138334

碩士 === 國立臺北科技大學 === 環境規劃與管理研究所 === 92 === The purpose of this study is to construct the “machine vision theory” to determine the occurrence of sediment hazards. Several digital image processing methods were developed under various sediment hazards floating scenarios. Compared to the traditional sensors, the image process of machine vision for detecting the occurrence of debris flow has the advantages of non-contacts and re-use. According to the analyses of the detecting system, the judgment can be real time and can be provided to the disasters prevention. The frames of this study including two main section. One is the image enhancement preprocessing and the other is image recognizing methods. The former was used to enhance blurred images, and the latter was to monitor the sediment hazards occurrence. In this study, the image preprocessing contained “local intensity equalization” and some useful filters. And image recognizing methods were focus on the scenarios of floating objects and image characteristics for sediment hazards. In order to calibrate the recognition thresholds and verify experiment results, a programming language, Borland C++ Builder, was chosen to develop several digital image processing programs. These calibrations were be analyzed by using documentary films, field images, and experimental images. In this study, many vague images were collected in the field and laboratory. The images of sediment hazards captured by the CCD camera were installed into computer. Image enhancement processing being done, the images became clear significantly. The 3D displacement measurements for a specific object by using machine vision theory were also analyzed herein. The errors of the measurements were around 6.1%. Having been reified by experiments, the specific object painted in white may be easily detected. When the difference of average gray level between the ROI of specific object and its background was low to 25, whether the specific object was within the ROI can be determined. In image texture, Fish’s linear discrimination function was applied to classify the debris flow and flood images. A superior recognition rate, near 90% in this study, was achieved. Hopefully the machine vision motoring system can be practically applied to the debris flow hazard mitigation in the further.