A Multi-feature Combination Method for Tracking of Marine Ships

Abstract

Maritime images and videos provide a variety of kinematic traffic information, including traffic volume, ship speeds and headings, which significantly benefits remote maritime traffic management and maritime safety improvement. To track multiple moving objects robustly from the complex original maritime image sequence, we propose an ensemble ship tracking framework with multiple feature extraction and Dempster–Shafer theory of evidence. Spatial features such as statistical features of gray values and geometric features in the image sequence rather than the motion characteristics are extracted, and the D-S theory is applied for combination of feature evidence to perform the correct association of multiple objects. We demonstrate the performance of the proposed ship tracking algorithm on a typical marine traffic scene through the marine surveillance video. Experimental results show that the proposed method can achieve continuous tracking of multiple marine moving objects with strong stability, and have the potential to improve marine transportation and fishery safety.