Object-based Moving Vehicle Extraction from Worldview2 Imagery

Moving vehicle detection is very important for transportation management and traffic monitoring. Due to the submeter spatial resolution of very high resolution (VHR) imagery, vehicles can be identified from this type of imagery. Furthermore, because of the slight time difference between image acquisition of onboard sensors, (i.e. Pan and MS sensors) in VHR satellite such as Quickbird and GeoEye-1, a moving vehicle is observed, by the satellite, at two different locations. Consequently, moving vehicles can be distinguished from the stationary ones by applying a proper change detection algorithm. WorldView2 possess three sensors, i.e. a Pan and two MS sensors (MS1 and MS2). Therefore, a moving vehicle is observed at three different locations. This feature together with the new spectral bands of WV2 adds opportunity to improve moving vehicle detection and extraction. This paper, utilizing an object-based framework, compares the automatic moving vehicle extraction by using the three pairs of WV2 sensors (i.e. Pan-MS1, Pan-Ms2 and MS1-MS2). The results show that of three image pairs, the MS1-MS2 is the best choice for moving vehicle extraction because of the larger time lag between MS1 and MS2 than between the Pan and MS1 or MS2.