Object Detection in Image with Complex Background

Dong Li, Yali Li, Fei He, Shengjin WangAdvances 2013 · DOI

Abstract: Object detection is the key technology in computer vision, with broad application prospects. Object detection has great research value and practical significance as a hot spot of video surveillance in recent years. The ship is a kind of important marine transportation, and ship detection is widely used in many fields. For example, government and port enterprises focus on the security management and risk control of dangerous ships in port operations. In addition, ship detection plays an important role in counting the number of ships and port throughput evaluation. In recent years, with the development of intelligent video surveillance technology, robust detection in complex backgrounds has become a priority.

Introduction

Detecting ships in harbor environments is notoriously difficult due to "background clutter." Unlike ships in the open ocean, harbor images contain wharfs, buildings, islands, clouds, and waves that can easily be mistaken for vessels. Traditional "sliding window" approaches often fail in these scenarios because they generate an excessive number of false alarms. This research proposes a system designed to distinguish between land and sea effectively, handling ships that are partially obscured or docked against complex shorelines, while maintaining the speed necessary for practical surveillance.

Methodology

The proposed framework utilizes a three-part pipeline to systematically narrow the search space and identify targets accurately.

The proposed multi-stage ship detection framework.

1. Sea-land Separation

To avoid searching the entire image, the algorithm first identifies the water. It employs several texture descriptors based on grayscale histograms and gray level co-occurrence matrices combined with a region growing method. Starting from a "seed" pixel known to be water, the algorithm expands to adjacent pixels with similar textures until it reaches the shoreline. This texture-based approach is more robust to lighting changes and waves than simple color-based methods.

2. Offshore Ship Detection

Once the sea is isolated, the system uses the Otsu algorithm—an adaptive thresholding technique—to find bright objects against the darker water. To filter out noise like whitecaps or small reefs, shape analysis is applied to check if candidate objects meet specific criteria for area, aspect ratio, and "duty ratio" (the proportion of the bounding box filled by the object).

3. Docked Ship Detection

Ships docked at wharfs are the most challenging to detect because their superstructures often blend into the land. The authors focus on the sea-land border area and utilize Chamfer matching. Instead of searching for a whole ship, the system looks for two specific "bottom boundary fragments"—the hull's shape where it meets the water. This multi-template approach is resilient to different ship types and partial occlusions.

Hull fragment template matching strategy for docked vessels.

Experimental Results

The algorithm was evaluated using a custom dataset of satellite images captured from Google Earth, consisting of 484 offshore images (497 targets) and 165 docked images (165 targets).

  • Offshore Detection: Achieved a 91.75% detection rate with an average processing time of 180ms per image. False alarms were low, averaging 0.15 per image.
  • Docked Detection: Achieved an 81.71% detection rate. While computationally heavier at approximately 1.5s per image, it successfully identified ships typically "invisible" to simpler algorithms.
  • Texture Performance: Contrast features based on gray histograms provided the optimal balance of discrimination and low computational cost.

Conclusion and Future Work

The research demonstrates a robust balance between accuracy and speed through hierarchical filtering. However, limitations remain; reefs and wharfs stretching into the sea can still trigger false positives due to shape similarity.

Future work will focus on incorporating inter-frame correlation for video surveillance. By analyzing motion information across multiple frames, the system can further reduce false alarms—since wharfs remain stationary while ships move—and increase overall processing speed.

Summary updated

Abstract

Object detection is the key technology in computer vision, with broad application prospects. Object detection has great research value and practical significance as a hot spot of video surveillance in recent years. The ship is a kind of important marine transharboration, and ship detection is widely used in many fields. For example, government and port enterprises focus on the security management and risk control of dangerous ships in port operations. In addition, ship detection plays an important role in counting the number of ships and port throughput evaluation. In recent years, with the development of intelligent video surveillance technology,
BibTeX
@inproceedings{li2013object,
  author = {Dong Li and Yali Li and Fei He and Shengjin Wang},
  title = {{Object Detection in Image with Complex Background}},
  booktitle = {Advances},
  year = {2013},
  doi = {10.2991/icmt-13.2013.58},
}
Figure