Batch Estimation‐Based Tracking Initialisation for Maritime Targets Using a Monocular Camera on Unmanned Maritime Vehicles
Yonghoon Cho, Sim Yong Lee, Jonghwi KimABSTRACT
This letter introduces an advanced state estimation initialisation framework for camera‐based Kalman filters to enhance maritime obstacle tracking. Exploiting the temporal data association of deep simple online and realtime tracking (DeepSORT), the proposed method effectively aggregates initial object detections from sequential camera images prior to confirming target registration. A batch estimation technique is then deployed on these detection sets to compute the optimal initial states, while the corresponding covariance matrices are analytically derived using the posterior Cramer–Rao lower bound (PCRLB). The practical efficacy and feasibility of this approach are demonstrated through comprehensive performance analyses on real‐world maritime datasets.