Year: 2024
Type: article
Abstract: Underwater object tracking holds considerable significance in the field of ocean engineering. Additionally, it serves as a crucial component in the operations of autonomous underwater vehicles (AUVs), particularly during tasks associated with capturing marine organisms. However, the attenuation and scattering of light result in shortcomings such as poor contrast in underwater images. Additionally, the motion deformation of marine organisms poses a significant challenge. Therefore, existing tracking algorithms face difficulty in direct application to underwater object tracking. To overcome this challenge, we propose a novel tracking architecture for the marine organism capturing of AUVs called ULOTrack. ULOTrack is based on a performance discrimination and re-detection framework and constitutes three modules: (1) an object tracker, which can extract multi-feature information of the underwater target; (2) a multi-layer tracking performance discriminator, which serves the purpose of evaluating the stability of the current tracking state, thereby reducing potential model drift; and (3) lightweight detection, which can predict the candidate boxes to relocate the lost tracked underwater object. We conduct comprehensive experiments to validate the efficacy of the designed modules. Finally, the results of the experimentation demonstrate that ULOTrack significantly outperforms existing approaches. In the future, we aim to carefully scrutinize and select more suitable features to enhance tracking accuracy and speed.
DOI: 10.3390/jmse12112092
Authors: Ju He, Yang Yu, Hongyu Wei, Hu Xu
Institution: Northwestern Polytechnical University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDG: Life below water
Open Access status: gold
Estimated APC paid: $2,955
Awards: G5076610014, G5791508200
OpenAlex ID: W4404502117