PSR-Based Feature Adaptation with Anomaly Re-Prediction for Maritime Target Tracking

The tracking method for the relevant filter cannot track when the target is affected by the background and in the scene of occlusion. We propose a target tracking algorithm based on the peak-to-sidelobe ratio (PSR) feature weight adaptation with anomaly re-detection based on the efficient convolution operator for manual features (ECO_HC) algorithm. The weights of the HOG features and CN features corresponding to the peak-to-sidelobe ratio are used as dynamic weight coefficients, and the maximum response value after their fusion is the area where the target may appear. Determine the target tracking status by calculating the tracking confidence of the response map, and when there is a tracking anomaly, the ECO_HC tracker is enhanced using the Kalman filter to re-predict the target position. The proposed algorithm has experimented on the OTB-50 dataset. The results show that our algorithm can cope with complex scenarios and improves by 3.0% in tracking accuracy and 2.8% in tracking success rate compared with the baseline algorithm ECO_HC. Finally, conduct tracking experiments on the actual sea surface scenes, and the proposed algorithm effectively suppresses the effects of common interference factors on the tracking effect.