摘要
针对传统HOG特征的行人检测方法中因遮挡及复杂环境存在较高漏检误检情况,建立了一种基于HOG和局部自相似(LSS)特征融合的行人检测算法。利用LSS反映图像内在几何布局和形状属性的特性,用主成分分析(PCA)将HOG和LSS两类特征在实数域降维,再将两种特征组合成新特征,结合线性SVM分类器进行行人检测。实验采用INRIA数据库和Daimler数据库作为训练集训练SVM,用730幅监控视频帧图片作测试集,将该方法与基于传统HOG特征的行人检测方法做对比,结果表明该方法平均漏检误检率降低16%,检测效果优于基于传统HOG特征的行人检测方法。
Aiming at the higher misdetection and error detection due to occlusion and complex environment in pedestrian detection based on traditional HOG features,this paper puts forward a kind of a pedestrian detection algorithm based on a fusion of HOG and local self-similar feature（ LSS）. it uses principal component analysis（ PCA） to reduce dimensionalities of traditional HOG feature and self similar（ LSS） feature which reflects the image inherent geometric layout and shape properties in the real domain respectively,then combines the two features into a new feature. Finnally,it builds a linear SVM classifier for pedestrian detection. The experiments uses INRIA database and Daimler database as training set for SVM,and chooses 730 images from a surveillance video as testing set. Comparing the method in this article to the pedestrian detection method based on traditional HOG feature,the results show that the average detection and error detection rate of the method in this article is reduced by 16%,better than the detection method for pedestrian detection based on traditional HOG features.
出处
《微型机与应用》
2016年第8期37-39,43,共4页
Microcomputer & Its Applications
