摘要
针对水面船舶的多目标实时定位,提出了一种定位和运动参数求解算法,采用固定位置和视角的单目摄像机采集船舶图像,对采集的图像进行高斯滤波和图像畸变校正,提出了基于船舶颜色、尺寸与运动学特征同时识别多个船舶目标(每个船舶目标独立识别)的方法,构建了图像坐标与真实环境坐标的转换模型与实时航速、航向和轨迹的计算模型,在水池环境下搭建了实时定位系统,开发了实时定位程序,并验证了定位方法的定位精度和轨迹跟踪性能。验证结果表明:在存在外界干扰的情况下,定位算法能实现对2艘船舶的精确识别;修正前坐标点横、纵坐标平均误差分别为0.058、0.209m,修正后分别为0.038、0.124m;摄像机定位数据更新频率为8 Hz,满足船舶控制需要;算法能实现对船舶位置、航速和航向的准确、实时计算,轨迹平滑且未出现异常点。
Aiming at the multi-target real-time locating of surface ship,an algorithm for real-time locating and motion parameters calculating of ship was proposed.The proposed algorithm was realized by processing the images captured by monocular camera with fixed position and perspective. The captured images were preprocessed with Gaussian filter and distortion correction,and a multi-target recognition method was proposed based on ship color features,size characteristics,and kinematic characteristics（every target could be identified individually）.A transformation model between image coordinate system and real coordinate system,and a computing model for ship speed,course and trajectory calculation were built.A real locating system was built in experimental pool,a real-time locating program was developed,and the locating precision and trajectory tracking performance were verified.Verification result indicates that in the circumstance with disturbances,the precise recognition of two ships can be realized by using proposed locating algorithm.The average locating errors in lateral and longitudinalorientation are 0.058 m and 0.209 mrespectively before amendment,0.038 m and 0.124 m respectively after amendment.The update frequency of camera locating data is 8 Hz,which can meet the control requirements.The real-time position,speed and course of ship can be correctly calculated,and the trajectory is smooth without any abnormal location point.1tab,17 figs,23refs.
出处
《交通运输工程学报》
EI
CSCD
北大核心
2015年第5期91-100,共10页
Journal of Traffic and Transportation Engineering
基金
国家自然科学基金项目(61273234
51479155)
交通运输部信息化技术研究项目(2013-364-548-200)
湖北省自然科学基金项目(2013CFA007)
关键词
船舶定位
单目视觉
多目标
颜色特征
畸变校正
目标提取
ship locating
monocular vision
multi-target
color feature
dostortion correction
target extraction
