Inferring Traffic Activity from Optical Satellite Images

In this paper we describe an approach to automatically estimate movements of vehicles in optical satellite imagery. The approach takes advantage of the fact that the optical axes of the panchromatic and multispectral channels of current spaceborne systems like IKONOS or Quickbird are not coinciding. The time gap that appears between the acquisition of the panchromatic and multispectral data can be used to derive velocity information. We employ a sub-pixel matching approach relying on gradient directions followed by least-squares fitting of Gaussian kernels to estimate the movement. The incorporation of the least-squares framework provides the basis to conclude about the accuracy of the movement estimates and to apply a statistical test deciding whether an object moves at all. We illustrate the matching and estimation scheme by various examples of real data. * Corresponding author.