Self-Selective Correlation Ship Tracking Method for Smart Ocean System

Xu Kang, Bin Song, Jie Guo, Xiaojiang Du, Mohsen GuizaniarXiv 2019 · arXiv

Introduction

Maritime surveillance is a critical component of "Smart Ocean" systems, yet tracking ships at sea presents unique challenges for computer vision. Traditional Correlation Filter (CF) methods often suffer from the boundary effect, where circular shifts used for training create unrealistic samples, leading to model drift. Furthermore, ships frequently undergo significant changes in scale and aspect ratio as they maneuver, which standard fixed-ratio trackers fail to capture. To address these issues, researchers have developed the Box Regression Correlation Filter (BRCF), a self-selective tracking algorithm designed for high accuracy and real-time performance in complex maritime environments.

Methodology

The BRCF framework introduces three primary innovations to enhance tracking robustness and efficiency.

Hard Negative Samples Mining

To mitigate the boundary effect, the model moves away from using all circular shifts. Instead, it selectively mines "hard" negative samples from the local search region. By focusing on samples that are difficult for the classifier to distinguish from the target, the model strengthens its discriminative power against complex backgrounds.

Illustration of hard negative sample selection for classifier training.

Adaptive Multi-Feature Fusion

The tracker utilizes a combination of diverse features: Histogram of Oriented Gradients (HOG) for shape, Color Histograms (CH) for appearance, and Local Binary Patterns (LH) for texture. The integration of these features is governed by an adaptive mechanism using Kullback–Leibler (KL) divergence. By measuring the reliability of each feature's prediction in real-time, the system dynamically adjusts their weights, favoring the most reliable feature under current environmental conditions (e.g., prioritizing HOG over CH in cases of severe glare).

The adaptive multi-feature fusion process, showing the integration of HOG, CH, and LH features for robust tracking.

Two-Step Scale Estimation

Accurate bounding box adjustment is achieved through a two-step process:

  1. Pre-estimation: Speeded Up Robust Features (SURF) keypoint matching between consecutive frames provides an initial estimate of scale changes.
  2. Box Regression: A dedicated regression model fine-tunes the bounding box, allowing for independent adjustments to width and height. This enables the tracker to adapt to changes in the ship's aspect ratio as it turns or tilts.

Experimental Setup

The BRCF method was evaluated using a Marine Traffic Dataset consisting of 10 to 20 high-definition (1920 × 1080) video sequences captured at 25 FPS. These sequences represent diverse and challenging scenarios, including ships entering or exiting bays, background interference from waves, and significant variations in distance from shore-based and ship-borne cameras.

Results and Discussion

Comparative analysis against industry-standard trackers, specifically Kernelized Correlation Filter (KCF) and Discriminative Scale Space Tracking (DSST), demonstrated the superiority of the BRCF approach.

  • Accuracy and Precision: BRCF achieved a success rate and precision over 8 percentage points higher than DSST on the marine traffic dataset.
  • Real-Time Performance: On a standard Intel i7-4790 CPU, BRCF reached an average processing speed of 44.98 FPS, nearly 22 FPS faster than DSST. This efficiency stems from the sparse selection of hard samples, which reduces the computational load.
  • Robustness: Qualitative evaluations showed that BRCF maintains a stable lock on targets even during long-range movement or heavy background clutter, where KCF and DSST often fail.

The use of KL divergence for reliability estimation proves highly effective for handling dynamic maritime environments. Furthermore, breaking the "rigid box" constraint through regression allows the model to accurately track the ship's silhouette throughout various maneuvers.

Conclusion

The BRCF algorithm provides a significant advancement in maritime ship tracking by combining selective sample mining with adaptive feature fusion and flexible scale estimation. While highly effective, the method's reliance on SURF keypoints suggests potential limitations in extremely low-visibility conditions (e.g., heavy fog). Future work will focus on evaluating the system on larger, global datasets and optimizing the algorithm for deployment on specialized edge hardware like NVIDIA Jetson modules for use in autonomous maritime platforms.

Summary updated

Abstract

In recent years, with the development of the marine industry, navigation environment becomes more complicated. Some artificial intelligence technologies, such as computer vision, can recognize, track and count the sailing ships to ensure the maritime security and facilitates the management for Smart Ocean System. Aiming at the scaling problem and boundary effect problem of traditional correlation filtering methods, we propose a self-selective correlation filtering method based on box regression (BRCF). The proposed method mainly include: 1) A self-selective model with negative samples mining method which effectively reduces the boundary effect in strengthening the classification ability of classifier at the same time; 2) A bounding box regression method combined with a key points matching method for the scale prediction, leading to a fast and efficient calculation. The experimental results show that the proposed method can effectively deal with the problem of ship size changes and background interference. The success rates and precisions were higher than Discriminative Scale Space Tracking (DSST) by over 8 percentage points on the marine traffic dataset of our laboratory. In terms of processing speed, the proposed method is higher than DSST by nearly 22 Frames Per Second (FPS).
BibTeX
@misc{kang2019self,
  author = {Xu Kang and Bin Song and Jie Guo and Xiaojiang Du and Mohsen Guizani},
  title = {{Self-Selective Correlation Ship Tracking Method for Smart Ocean System}},
  year = {2019},
  eprint = {1902.09690},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1902.09690},
}
Figure