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：2006
卷号：XXXVI Part 3
出版社：Copernicus Publications
摘要：Vehicle detection is motivated by different fields of application, e.g. traffic flow management, road planning or estimation of air and noise pollution. Therefore, an algorithm that automatically detects and counts vehicles in air- or space-borne images would effectively support these traffic-related analyses in urban planning. Due to the small vehicle size in satellite images detection of single vehicles would deliver ambiguous results. Hence, our scheme focuses primarily on the extraction of vehicle queues, as the pattern of a queue makes it better distinguishable (as a whole) from similar objects. Hypotheses for queues are generated by sophisticated extraction of ribbons. Within these ribbons single vehicles are searched for by least-squares fitting of Gaussian kernels to the width and contrast function of a ribbon. Based on the resulting parameter values, false and correct hypotheses are discerned. The results show that the analysis of width and contrast information using least square optimization is able to extract single vehicles from queues with high correctness. Still, the completeness of the overall extraction is relatively low, since only queues can be extracted but no isolated vehicles. The results clearly show that the approach is promising but further improvements are necessary to achieve a higher completeness
