期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2007
卷号：XXXVI-5/C55
出版社：Copernicus Publications
摘要：Captured imagery by terrestrial mobile mapping systems proved to be an effective tool for the extraction of road and road-side features for a variety of transportation activities. Among these features, road signs are important for several applications such as navigation and driver's alert systems. Manual extraction of road signs from thousands of imagery is a ver y expensive and tedious process. To mitigate such a problem, this paper proposes a methodology for the generation of hypotheses regarding instances of road signs in the captured imager y. Such a process would reduce the manipulated imagery from several thousands to few hundreds. The task of object recognition involves segmenting and extracting regions and categorizing the regions into predefined classes. Inherently, this is an ill-posed problem. Therefore, a variety of attributes of the extracted regions need to be incorporated for better analysis. Spectral information is one such important attribute. In the case of color imagery, color content makes up the spectral information. In addition to the spectral information, geometric characteristics of the imaging system and the sought-after objects play a significant role in regularizing the recognition process. More specifically, the proposed strategy will start by using the spectral information of the sought-after signs to isolate regions in the captured imagery with similar signatures. Then, the geometric characteristics of the imaging system will be used to trim some of the defined regions. In other words, the imaging configuration and the driving trajectory relative to the road will be used to define an area of interest where road signs are most probably located. Finally, geometric attributes of the defined regions (such as size, regularity of the region boundaries, and moments) will be used to generate a hypothesis regarding an instance of a road sign. The performance of the proposed methodology will be evaluated by experimental results involving real datasets captured by a terrestrial mobile mapping system
