#38: Hierarchical motion decomposition for cloud-tracking

C. Kambhamettu, K. Palaniappan, and A. Hasler

17th Int. AMS Conf. on Interactive Information and Processing Systems (IIPS) for Meteorology, Oceanography and Hydrology, pgs. 318--323, 2001

motion, tracking, parallelization, cloud, remote sensing

PlainText, Bibtex, URL, Google Scholar


Automatic cloud tracking is an extremely important and challenging problem. Tracking clouds provides an estimate of the cloud-drift wind velocities, which is useful for a number of meteorological and climate applications. In this work, we propose a novel method for automatic tracking of clouds, including the tracking of eye of the storm in case of hurricanes. To the best of our knowledge, this is the first integrated system for automatic cloud tracking using super-rapid scan data which produces a dense wind-field estimate using a hierarchy of motion models that includes global storm motion tracking. We have experimented with 490 frames of a remarkable 1-minute image timeseries of hurricane Luis, with a dynamic resolution of 10 bits per pixel captured by the GOES-9 satellite. Our approach uses hierarchical motion decomposition technique, wherein the first level models the global rigid motion and tracks the eye of the hurricane (thus the storm motion), the second level models the local non-rigid deformable motion and tracks the storm-relative clouds at a coarse scale using region-based image subtraction, and the third level models the local non-rigid semi-fluid motion and tracks storm-relative deforming clouds at a finer scale using region based cross-correlation. Output from the system includes accurate velocities with the latitude, longitude and time information for each pixel in the hurricane sequence. We demonstrate a high accuracy of the algorithm through our experiments using comparisons with manual tracking analysis. Visualization of the results is performed using the Interactive Image Spread Sheet, designed for such hyper-image environments. Future work includes the use of multispectral information, cloud motion segmentation and developing physical models for operational cloud-tracking performance.