Long-term correlation robust tracking of visual targets for unmanned surface vehicles using multi-feature fusion

Objective To address the problem of visual target tracking failure caused by significant wave interference and severe camera shaking in unmanned surface vehicles (USVs), a multi-feature fusion long-term correlation robust tracking algorithm is proposed.

Methods First, the multi-feature fusion technique is used to enhance the expression of target features and improve the robustness of the target model. Then, high-dimensional feature dimension reduction and response map sub-grid interpolation are utilized to improve the efficiency and accuracy of target tracking. After that, a mechanism for water surface target re-identification is designed to address the issue of stable tracking when the target is completely out of sight. Finally, the proposed algorithm is validated and compared through multiple representative video datasets.

Results  The experimental results show that compared with traditional long-term correlation tracking algorithms, the average success rate is improved by 15.7%, the average distance precision index is improved by 30.3% and the F-score index is improved by 7.0%.

Conclusion The proposed algorithm can handle target tracking failure in harsh marine environments and has important technical support significance for improving the intelligent perception capability of USVs and ocean robots.