Combining Float Car Data and Multispectral Satellite Images to Extract Road Features and Networks

This chapter presents an automatic methodology for the extraction of spatial road features and networks from floating car data (FCD) that was integrated with multispectral remote sensing images in metropolitan areas. This methodology is divided into two basic steps. Firstly, a spatial local statistical examination is carried out to extract the nodes of each road segment. Based on the local Moran’s I statistics, a new statistic method is developed to detect local clusters. Significance is assessed using a Monte Carlo approach to determine the probability through observing large samples under the null hypothesis of no pattern. When all the necessary nodes are detected, spatial road segments can then be organized by linking pairs of nodes, which are used as the candidate road segments for the next step. Secondly, pre-processed multispectral remote sensing images are utilised for testing those initial road segments. To prove the concept, a Metropolitan area is employed as a case study. Road segments with high significance values in the tests are selected to construct the spatial road network. The developed methodology could be adopted for the provision of high quality navigational road maps in a cost-effective manner and the experimental results are presented.