摘要
行人检测是智能监控技术的一项重要内容,为了快速准确地对地铁站内的行人进行检测,采用图像的梯度向量直方图(Histograms of Oriented Gradient,HOG)特征表征行人,并通过改进HOG特征的提取算法,加快了特征向量的提取速度。分类器使用支持向量机(SVM),针对大量行人和背景训练样本,提取HOG特征并训练SVM行人分类器。用训练得到的分类器对测试样本进行检测,实验表明,提出的方法具有较高的识别率和较强的普适性,并可以应用于实际地铁环境中。
In order to have a rapid and accurate detect of the pedestrian in metro station, the characteristics of histograms of oriented gradient (HOG) are used to characterize pedestrian. By improving extraction algorithm with HOG characteristics the extraction speed of eigenvector is accelerated. Classifier uses support vector machine (SVM). According to a large amount of pedestrian and background training samples, the HOG characteristics are extracted, and the SVM pedestrian classifier is trained. The tests show that the proposed method has higher recognition rate and stronger universality, which can be used in practical metro environment.
出处
《现代城市轨道交通》
2010年第2期31-33,36,共4页
Modern Urban Transit
基金
国家自然科学基金项目(60973061)
