Motion Detection from Satellite Images Using FFT Based Gradient Correlation

In this paper, we study the problem of quantifying target motion at the Earth’s surface. We leverage satellite images with different times to detect the movement of different targets. We focus on the motion detection of vehicle and exploit the near-simultaneous satellite images to assess the vehicle trajectories. We segment the target images into fixed size blocks and use FFT based gradient correlation to determine the displacements of each block. Then we reduce the block size and utilize an iterative multigrid image deformation method to calculate the global velocity field and improve the accuracy of motion detection. Compared to other correlation method, the FFT based gradient correlation is more accurate and time efficient, which can estimate translations, arbitrary rotations and scale factors. Moreover, the use of image gradient is able to capture the structure feature of salient image and make the correlation more robust. We do our experiments using images acquired by Planet Dove Satellites. The experiments show that our algorithm can quantify target motion robustly and efficiently. Our algorithm enhances the ability of time series satellite images to be used for motion detection.