基于形状特征的移动目标实时分类研究  

Moving target classification based on shape features from real-time video

作　　者：侯北平[1] 朱文[1] 马连伟[1] 介婧[1] 

机构地区：[1]浙江科技学院自动化与电气工程学院,杭州310023

出　　处：《仪器仪表学报》2010年第8期1819-1825,共7页Chinese Journal of Scientific Instrument

摘　　要：基于室外场景中移动目标的形状特征,提出了一种利用多级神经网络进行运动目标实时分类的方法。在对人、人群、自行车、大汽车与小汽车5种常见运动目标的形状特征分析基础上,将七阶矩、分散度、长宽比、离散度作为分类特征向量,利用多级神经网络进行了目标分类。实验结果表明该方法误判率低、分类精度高,能较好地实现单人、人群、自行车、汽车的分类。This paper presents a moving target classification algorithm based on shape features and multi-stage neural network architecture.The shape features of human,human group,bike,bus and car are analyzed and studied.Moment,dispersedness,compactness and length-width ratio are used as the classification eigenvectors.The detected video objects can be categorized into pre-defined groups of human,human group and vehicle by using multi-stage neural network.Experiment results from test images and real-time video are provided,which demonstrate the effectiveness of the proposed method.

关 键 词：人车识别 运动检测 神经网络 目标识别 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]