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：Inventory of road infrastructure represents a key application for integrated mobile mapping systems. The classical approach is to post-process geo-referenced imagery that has been captured from mobile mapping vans while driving within the road. Problems arise then within city centres where dense traffic and parking vehicles may often hinder occlusion free image captures of the objects of interest. We present an alternative solution in terms of a mobile PDA based geo-information system for pedestrian based road sign data collection which is targeted towards complementing the classical approach, particularly in city centres. The novel system consists of a mobile computing unit which is equipped with a GPS sensor, a digital camera, a standard mobile GIS software, and a user interface for minimal user intervention. Imagery for road sign inventory is firstly captured and geo-referenced in the street and then analyzed in a second, off-line processing step. We present a post-processing tool for semi-automated classification of road signs, and in parallel an identification of the sub-plate information, strongly supporting the processing of huge data sets, e.g., for the processing of data from complete city areas. The tool includes advanced image analysis methodology with (i) localization of selective key descriptors and regions of interest, (ii) the fitting of geometrical constraints to the extracted set of interest descriptors, and (iii) the matching of content information from the visual information within the sign plate. Text information on the sub-plate is first detected using a wavelet based texture filter, and then post-processed with segmentation operators and standard OCR software. The system has been successfully applied in a project on a complete inventory within the City of Graz, Austria, processing more than 30.000 road signs. The system achieved high accuracy (≈90%) in the automated image based classification of road signs on more than 20 different road sign classes
