张恒　杨家轩　周洋宇　姜苗苗　王毓玮

摘要：

为降低海事监控视频图像背景中运动物体引起的杂波和噪声对船舶目标检测的影响，根据采集的可见光视频图像特性，提出一种海天背景下船舶目标自适应检测算法。将待检测图像进行预处理，使用自适应中值滤波和均值漂移（mean-shift）滤波对图像进行滤波去噪。采用密度峰聚类对传统K均值聚类算法进行改进，自适应确定初始聚类中心及其数量。对海面船舶进行自适应聚类分割。仿真实验显示：该算法的检测准确率为90.3%，验证了其准确性和可靠性;单帧视频图像的船舶目标检测用时可控制在100 ms以内，满足实时检测的要求。结果表明：该算法可以实现海天背景下船舶目标的准确、快速检测，为海上船舶目标跟踪奠定了可靠的基础。

关键词：

交通工程; 船舶目标检测;K均值聚类; 密度峰聚类; 自适应中值滤波; 均值漂移滤波

中图分类号：  U675.79; TP391.4

文献标志码：  A

收稿日期： 2020-04-20

修回日期： 2020-08-20

基金项目：

国家自然科学基金（51579025）;辽宁省自然科学基金（20170540090）

作者简介：

张恒（1996—），男，安徽亳州人，硕士研究生，研究方向为海上运动目标检测，（E-mail）dlmuzhangheng@163.com;

杨家轩（1981—），男，山东鱼台人，副教授，博士，研究方向为海上交通信息工程，（E-mail）yangjiaxuan@dlmu.edu.cn

Adaptive detection of ship targets in visible light video images

ZHANG Henga，b， YANG Jiaxuana，b， ZHOU Yangyua，b，

JIANG Miaomiaoa，b， WANG Yuweia，b

（a.Navigation College; b.Key Laboratory of Navigation Safety Guarantee of Liaoning Province，

Dalian Maritime University， Dalian 116026， Liaoning， China）

Abstract：

In order to reduce the influence of clutter and noise caused by moving objects in the background of maritime surveillance video images on the detection of ship targets， according to  the characteristics of the collected visible light video images， an adaptive detection algorithm for ship targets in the sea-sky background is proposed. The images to be detected are preprocessed， that is， the adaptive median filtering and the mean-shift filtering are used to filter the images to remove noise points. The traditionalK-means clustering

algorithm is improved by the density peak clustering， and the initial cluster centers and its number are adaptively determined. The adaptive clustering is used to segment ships on the sea. The simulation experiment  shows that， the detection accuracy of the algorithm is 90.3%， which verifies its accuracy and reliability; the ship target detection time of an image can be controlled within 100 ms， which can meet the requirement of real-time detection. The results show that， the algorithm can achieve accurate and rapid detection of ship targets in the sea-sky background， and lays a reliable foundation for tracking ship targets at sea.

Key words：

traffic engineering; ship target detection; K-means clustering; density peak clustering; adaptivemedian filtering; mean-shift filtering

0 引 言

海上船舶目標检测具有重要的现实意义和应用价值。在军事上，可以保障我国领土与主权完整并促进海域的良性管理等;在民用上，可以帮助海事管理人员对进出港船舶进行管理，减少船舶事故的发生，并对船舶非法行为进行更为有效的监控。……