基于轮廓特征的场景动态目标实时分类研究  

Research on Real-time Classification of Scene Moving Target Based on Contour Features

作　　者：彭晖[1] 刘士荣[1,2] 仲朝亮[1,2] 

机构地区：[1]杭州电子科技大学电气自动化研究所,杭州3100182 [2]检测仪表与自动化系统集成技术教育部工程研究中心,杭州310018

出　　处：《控制工程》2015年第3期393-397,共5页Control Engineering of China

基　　金：国家自然科学基金(61175093)

摘　　要：针对动态场景目标的分类问题,提出一种基于轮廓特征的运动目标分类方法。通过构建多种轮廓特征相结合的特征向量模型来描述动态目标,作为分类器学习判别的基础。该方法首先通过混合高斯模型提取出视频中的动态场景目标,经图像形态学的处理,获得较为精确的动态场景目标轮廓图像,然后使用特征向量模型提取轮廓的相关特征作为分类器学习判别的依据,并得到最终分类结果。以常见的运动目标汽车、行人、自行车作为分类类别进行实验。结果表明该方法有较高的分类精度,且具有实时性好,易于实现的特点。To deal with the problem of dynamic scene target classification, areal-time classification method for moving target based on contour features is presented. As the learning and classifying foundation of classifier, vector model combined with a variety of contour features is proposed to describe the moving target. Firstly, Gaussian mixture modelis used for motion detection. More accurate contour image of moving target is achieved after image morphology processing, and then contour features extracted by feature vector model are applied as the basis of classifier learning and classifying. Finally, classification results are obtained. Support Vector Machine （SVM） is employed to classify moving targets, such as humans, vehicles and bikes. Experiment results demonstrate that this approach has higher classification accuracy, better real-time performance, and is easy to implement.

关 键 词：动态目标分类 轮廓特征 特征向量模型 形态学处理 支持向量机 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]