Data association for autonomous ships based on virtual simulation environment

Abstract

Ships obtain information regarding their surroundings using various types of sensors such as automatic identification systems (AISs), radio detection and ranging (RADAR), and cameras. Each sensor data point includes different information and exhibits different characteristics, and the measurement data generated from a single target must be associated to increase the recognition accuracy and tracking consistency. Whereas AIS data include the maritime mobile service identity, which is a unique identifier for ships, the information of ships detected by RADAR or cameras is limited; thus, an identifier must be assigned through data association. In this study, an improved data-association algorithm, which uses the coherent point drift and nearest neighbor methods, is developed. Additionally, the data-association algorithm is verified using actual sensor data generated during ship navigation. A virtual simulation environment is constructed to generate sensor data for the development and verification of the data-association algorithm. Subsequently, target ships are detected from virtual cameras and RADAR images, and the visual information of the ships is extracted from the camera images. Subsequently, the measurement data are associated using the position, speed, and length of the target ships to improve the accuracy of the data association. Finally, the applicability of the proposed algorithm for data association is verified via a virtual simulation. Results show that the proposed algorithm can be applied to practical scenarios involving severe conditions for data association.