| A series of sensors on the pavements being used for traffic flow data detection have some disadvantages. With the advance of high-resolution commercial remote sensing system, detecting traffic flow data from remote sensing images at large regions has become an attractive option in the fields of traffic information collection technology. The main goal of the research is to explore the vehicle information from high-resolution satellite remote sensing images, and to further extract the traffic flow parameters. The study includes the following aspects: (1) A new method for vehicle detection based on object-oriented image analysis is developed and applied. By analysis of high-resolution satellite images characteristics, object-oriented segmentation is implemented to generate image objects, and then feature space is built by extracting features of these objects, which is used for vehicle detection and classification, furthermore for traffic flow information analysis. (2) Another method, based on support vector machine classification, is established to achieve the identification and classification of vehicles. According to the characteristics of vehicles in remote sensing images, the image texture feature is extracted to obtain texture feature images. In the texture feature images vehicle samples and non-vehicle samples are selected for training classifier. We set kernel parameters to the SVM classifier for vehicle detection; (3) From the results of vehicle classification, extraction of traffic flow parameters is discussed, including direct and indirect traffic flow parameters extraction; (4) Eventually, experiments and results are described and assessed, using QuickBird2 data. And it is proved that the object-oriented method is valid with the overall accuracy of 90%. However, there is still some error classification, and the shadows of trees and buildings are the main reason. Thus, it is necessary to eliminate shadows in the pre-processing of satellite images. This research provides a reference for traffic flow information collection in ITS when compared to traditions. |
