|
一种面向船舶智能航行的海上目标实时跟踪方法 |
A Real-time tracking method for ship Intelligent Navigation |
投稿时间:2019-07-25 修订日期:2019-09-04 |
DOI: |
中文关键词: 智能船舶 目标跟踪 相关滤波 高斯混合模型 因式分解卷积 |
英文关键词: Intelligent ship Target Tracking Correlation filtering Gaussian mixture model Factorization convolution. |
基金项目:国家自然科学基金项目(面上项目,重点项目,重大项目) |
|
摘要点击次数: 304 |
全文下载次数: 0 |
中文摘要: |
海上目标感知的准确性和实时性是实现船舶智能航行的前提和基础。为了满足以上要求,将有效卷积算子(Efficient Convolution Operators,ECO)引入到海上船舶目标的跟踪中。该算法以相关滤波为基础,响应值的最大值之处为目标船舶中心所在的位置。获得中心位置之后,采用尺度滤波方法估计出船舶目标的最佳尺度,从而完成对目标当前帧的跟踪。该算法利用因式分解卷积的方式分解卷积,降低数据维度,减少计算时间;采用高斯混合模型将样本分成不同的类别,降低训练集的样本冗余度;采用稀疏更新策略更新样本模型,防止过拟合问题。通过选取了海洋环境下的船舶不同运动场景作为实验样本,并与几种常用跟踪算法进行实验对比,验证了ECO算法在海上船舶目标跟踪上的准确性和实时性。 |
英文摘要: |
The accuracy and real-time performance of marine target perception is the premise and foundation of ship intelligent navigation.?In order to meet the above requirements, the effective convolution operator (Efficient Convolution Operators,ECO) is introduced into the tracking of ship targets at sea.?The algorithm is based on correlation filtering, and the maximum response value is the position of the center of the target ship.?After obtaining the center position, the scale filtering method is used to estimate the optimal scale of the ship target, so as to complete the tracking of the current frame of the target.?In this algorithm, the convolution is decomposed by factorization convolution, the data dimension is reduced and the calculation time is reduced, and the Gaussian mixture model is used to divide the samples into different categories to reduce the redundancy of the training set.?The sparse update strategy is used to update the sample model to prevent the over-fitting problem.?By selecting different ship motion scenes in marine environment as experimental samples and comparing with several commonly used tracking algorithms, the accuracy and real-time performance of ECO algorithm in marine ship target tracking are verified. |
View Fulltext
查看/发表评论 下载PDF阅读器 |
关闭 |