Efficient Association Multiple-Object Tracker for Marine Benthonic Organisms Counting

The number of existing marine benthonic organisms fundamentally affects the aquaculture decisions of enterprises, yet faces significant challenges in automatic counting in dynamic occluded environments. In this paper, to tackle the above challenges, an efficient association multiple-object tracker for marine benthonic organisms counting (EAMBOC) is innovatively proposed by combining a Kalman filter incorporating global motion compensation (KGMC) and an efficient association mechanism (EAM). Specifically, KGMC is devised to alleviate the effect of nonlinear camera motion on trajectory prediction by correcting the Kalman filter via the affine matrix. EAM is further employed to associate the predicted trajectories with the detected objects based on the confidence scores so that valid information can be mined from low-quality detections to maintain the continuity of the trajectory. Finally, a comprehensive comparison with typical schemes using real-world videos demonstrates that EAMBOC outperforms DeepSORT, ByteTrack and UAVMOT schemes in terms of RMSE, MAPE and $\mathrm{R}^{2}$ metrics.