01656968

Component

In conventional ship visual tracking methods, it is difficult to recognize the tracking ships when they intersect or sail in heavy traffic area as ships present different appearance and shape features in videos. To overcome this difficulty, the authors propose to employ Multi-view learning algorithm with multiple distinct feature sets to improve the effectiveness and the robustness of the ship tracking. First, the authors explore and exploit multiple distinct feature sets consisting of Laplacian-of-Gaussian (LoG) descriptor, local binary patterns(LBP) descriptor, Gabor filter, histogram of oriented gradients (HOG) descriptor and Canny descriptor, which respectively presents geometry structure, texture, contour information etc. Then, the authors propose a framework of ship tracking integrating Multi-view learning algorithm and sparse representation method to implement the robust and effective ship tracking. Finally, the methodology is evaluated over different vessel traffic scenarios. The performance of the proposed method has been improved in accuracy and robustness compared with conventional typical ship tracking methods.

This paper was sponsored by TRB committee ABJ50 Standing Committee on Information Systems and Technology. Alternate title: Robust Ship Tracking via Multiview Learning and Sparse Representation.

18-02553

English

8p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; Photos; References

Accuracy; Algorithms; Automatic tracking; Detection and identification systems; Methodology; Ships; Video

Data and Information Technology; Marine Transportation; Planning and Forecasting

Transportation Research Board Annual Meeting 2018 Paper #18-02553

TRIS, TRB, ATRI

Jan 8 2018 10:37AM