Automatic Ship Tracking by Flow Analysis on Radar Images

Abstract

This paper proposes a method to track ships automatically using B-scope RADAR images received from a remote RADAR station located on the coast of Tokyo Bay. The images are in time series, hence flow analysis can be applied for ship tracking. In this study, optical flow, which is distributions of moving vectors, was calculated to detect and track ships. Although there are many ways to compute optical flow, we chose the LUCAS-KANADE technique for processing since ships move slowly and steadily. The LUCAS-KANADE technique uses Gaussian Windows to predict the position of a ship in the next frame. The pyramid method was used in three steps to facilitate calculation. It is necessary to separate clutters from ship data because optical flow obtained through the processing contains both. This was achieved by using as the criteria migration length and the sum of the moving vectors of each ship. As a result of the processing, 75-85% of real ship were detected and tracked in normal condition and 60-70% under difficult circumstances. We also related the processing results to geographic and time information so that users could understand where and when ships passed.