An Initial Study on Vehicle Information Extraction from Single Pass QuickBird Satellite Imagery

Vehicle information is useful in many fields. In this paper, a technique is presented to extract vehicle information from single pass QuickBird images. While passing a target area, the satellite acquires panchromatic (PAN) and multi-spectral (MS) images simultaneously. Because of a small time interval difference between the PAN and MS images, it is theoretically possible to extract two sets of vehicle ground positions from the PAN and MS images, respectively, to identify whether or not a vehicle is in motion, and to calculate the vehicle’s velocity and direction. Practically, however, this extraction and calculation are challenging. Since the time interval is very short, a small error in information extraction will result in an unacceptably large error in the calculated vehicle position and velocity. Another challenge is that satellite image pairs (PAN and MS) do not have the same resolution. Therefore, traditional change detection techniques are incapable of providing reliable results, due to varying scales and relief distortions in the co-registered images or slight pixel shifts in the orthorectified images caused by resampling of the PAN and MS images. In order to avoid these errors, this research presents an algorithm through which a vehicle’s ground positions can be directly calculated from the raw PAN and MS images. Experiments demonstrate that it is feasible to use this technique to extract vehicle information from high-resolution images obtained from a single satellite pass.