摘要
用于船舶吃水线识别的机器视觉检测方法采用固定相机获取水尺区域,存在船舶外弦吃水线图像数据采集难度大、吃水线测定效率低等问题。基于此,利用无人机搭载相机航拍图像,充分考虑其图像特点,提出一种吃水线区域动态识别算法。无人机航拍图像克服了已有固定相机的不足,但是,由于无人机飞行航迹的变化导致吃水线图像存在波动,此外,吃水线区域的水迹线、水波纹以及不同曝光环境也会影响吃水线的识别精度,为此,首先采用融合水尺特性先验知识的轮廓统计筛选方法,提取先导知识,定位整幅图像中感兴趣的吃水线水平区域;进而综合利用像素点在L×a×b颜色空间的信息,采用知识引导的K-means++聚类和分水岭算法,实现吃水线的精确分割。面向多场景下的航拍图像,实验结果表明,所提算法可以有效避免光照条件和水迹线等的干扰,实现动态视频下船舶吃水线区域的快速识别,对港口这一复杂航拍场景具有较强的环境鲁棒性,为准确实时水尺计重提供有效保障。
In traditional image-based waterline identification method,a fixed camera is used to capture the waterline area,which requires complex equipments and has difficulty to measure the waterline of the outer chord of ships. This will decrease the identification efficiency. Based on this,a dynamic identification method of waterline area based on aerial image captured by UAV is proposed. Though UAV is more convenient than the fixed cameras,the waterline area in the images may be unsettled due to the unstable flight track. In addition,water trace,ripples in the surface of water and different exposure environment all have the adverse impact on the recognition of waterline. Thus,a statistic-based selection for contour combined with the prior knowledge of waterline characteristics is adopted to locate the interesting horizontal areas containing the waterline. After comprehensively utilizing the color and spatial information of pixels,the waterline is accurately segmented by knowledge-guided Kmeans++ and watershed algorithm under L ×a ×b color space. The experimental results show that the proposed method can effectively process the influence from light conditions,water traces and so on,as well as rapidly identify the waterline area from a dynamic video. Its strong robustness for the complex environment in ports provides the foundation for obtaining the accurate weigh of cargoes.
作者
程健
安鸿波
郭一楠
叶亮
CHENG Jian;AN Hongbo;GUO Yinan;YE Liang(Research Institute of Mine Big Data,China Coal Research Institute,Beijing 100013,China;School of Information and Control Engineering,China University of Mining and Technology,Xuzhou 221116,China;School of Electromechanical and Information Engineering,China University of Mining and Technology(Beijing),Beijing 100083,China)
出处
《郑州大学学报（工学版）》
CAS
北大核心
2021年第3期47-52,共6页
Journal of Zhengzhou University（Engineering Science）
基金
国家自然科学基金资助项目(61973305)
辽宁省自然科学基金资助项目(2020-KF-22-02)
中国煤炭科工集团有限公司科技创新创业资金专项重点项目(2019-2-ZD002)。
关键词
航拍图像
吃水线区域
轮廓统计筛选
知识引导
动态识别
aerial image
waterline area
statistic-based selection for contour
knowledge guidance
dynamic identification
