Street Sign Detection On Google Street View Images

• Main Authors: Jyun-Han Chen, 陳俊翰
• Other Authors: 蔡榮得
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/60501063620044898237

碩士 === 國立中興大學 === 土木工程學系所 === 104 === Conventional three-dimensional building modeling requires on site shooting for data gathering, therefore, it is not only time-consuming but also very expensive. In this study, we propose a method that utilizes Google Street View service and route planning by using JavaScript, CSS and HTML to build a web application which can also be used to retrieve raw image data. Python, an interpreted cross-platform programming language, is used to implement algorithms in filtering, edge detection, morphological image processing, and set constraints for automated street sign detection. In addition, parallel computing (multiprocessing) is adopted for the implementation of these algorithms that can be executed on Amazon EC2, a flexible deployment resources service for cloud computing. Finally, Flask, a web application framework, is used to distribute processed images from the back-end side to the front-end side. This fully automatic street sign detection with video streaming preview system can provide a wide range of subsequent applications. For a path with a total of 176 images, it takes 4.49 minutes to process, leading to 1.53 seconds for each image when using a 32-core computer for operations. The rent on Amazon EC2 at the time of this study is US $ 1.68 per hour. Detection accuracy of street signs is 80.56% among these test street view images.