[4P2-J-3] Data mining: spaciotemporal data

[4P2-J-3-04] Modeling and tracking spatiotemporal objects by mixture distribution

Application to phased-array weather radar data

Ryo Hayashi¹, Yuki Fuji¹, 〇Rie HOnda¹, Shinsuke Sato², Takefumi Murata², Kazuya Muranaga³, Kentaro Ukawa³, Koji Sasa¹, Fumie Murata¹ (1. Kochi University, 2. NICT, 3. SEC CO.LTD)

Keywords:multivariate normal distribution, mixture model, EM, spatiotemporal object, tracking

We propose a new method to detect and track spatiotemporal objects via mixture mode of multivariate normal distribution. Modified Greedy EM algorithm is used to obtain emerged and vanished components. As a post-processing, event-type such as fusion or separation are distinguished by comparing current parameters and previous parameters and each component is labelled to reflect the event. The method is applied to the analysis of phased array weather radar data analysis.