James N. K. Liu and Bo Feng, The Hong Kong Polytechnic University
Conventionally, the identification of Tropical Cyclones (TCs) has been highly dependent on subjective human judgment applied to vast amounts of data extracted from satellite images. The most popular approach to comparing two given TCs is to measure the distance between various contour points of the TC extracted from a satellite image. However, this measure has a very high computational cost as it involves point-to-point calculations. In this paper we present a novel approach to TC comparison that is based on a TC’s typical spiral shapes. Given two sets of contour points, one for each tropical cyclone, we adjust the similarity of two shapes using angle features found among the successive contour edge points. The adoption of a time warping approach results in a fast and accurate comparison. Implementation and evaluation makes use of a two-layer prototype. Experimental results show that our approach performs better than conventional methods such as the Hausdorff distance measure.