海上船舶智能检测及其关键算法研究

Research on Marine Ship Detection and Its Algorithms

海上船舶检测环境复杂、实时监测难度大。针对传统滑动窗口遍历搜索法运算量大、实时性差的问题,提出了一种基于似物性和相似度匹配的快速传播检测算法。首先,提取正、负样本的规范二值化梯度特征,用支持向量机的方法训练分类器,用于检测显著性区域。其次,针对分类器多尺度筛选的问题,引入相似度匹配的过滤算法。最后,针对船舶上人员的检测,运用方向梯度直方图(Histogram of Oriented Gradient,HOG)+支持向量机(Support Vector Machine,SVM)的方式训练分类器,提取候选框的HOG特征作为目标判定。

The marine ship detection environment is complex and the real-time monitoring is difficult. A fast propagation detection algorithm based on similarity matching and similarity matching is proposed to solve the problem of large computation and poor real-time performance of traditional sliding window traversal search method. Firstly, the canonical binarization gradient features of positive and negative samples are extracted, and then the classifier is trained with a Support Vector Machine method to detect significant regions. Secondly, to solve the problem of multi-scale classification, a similarity matching filtering algorithm is introduced. Finally, the Histogram of Oriented Gradient(HOG)+Support Vector Machine(SVM) is used to train the classifier, and the HOG feature of the candidate box is extracted as the target.