一种基于HOG和颜色自相似性特征的行人检测方法

龙源期刊网 http://www.qikan.com.cn

一种基于HOG和颜色自相似性特征的行人检测方法

作者：张金慧 吴斌 邵延华

来源：《电脑知识与技术》2018年第02期

摘要：行人检测在计算机视觉领域中越来越广泛的应用，使其有着重要的研究意义。尽管技术得到了显著的改进，行人检测仍然是一个存在挑战的难题，需要更精确更高效的算法。针对HOG特征的传统检测方法中存在的问题，该文提出一种融合颜色自相似性（CSS）特征的方法。利用颜色自相似性特征与描述人体轮廓特征的HOG特征互补，CSS反映图像内在几何布局和形状属性的特性，为提高检测效率，使用经主成分分析法（PCA）降维处理HOG和CSS特征。实验采用INRIA数据集作为训练样本训练SVM。对比单一使用HOG，该文的方法在检测速度和准确性上得到有效提高。实验结果也验证了本算法的有效性。 关键词：行人检测；梯度方向直方图；颜色自相似性；融合

中图分类号：TP391 文献标识码：A 文章编号：1009-3044（2018）02-0146-03 Method Human Detection Based on HOG and CSS Characteristics ZHANG Jin-hui， WU Bin， SHAO Yan-hua

（School of information and engineering， Southwest University of Science and Technology，Mianyang 621010， China）

Abstract： Pedestrian detection is becoming more and more widely used in the field of computer vision， and it has important research significance. Despite the significant improvements in technology， pedestrian detection is still a challenging problem， requiring more accurate and efficient algorithms. For the existing problems in the traditional detection of HOG， this paper proposes a method of fusing color self similarity （CSS） features. CSS is complementary to the HOG that describes the human contour features. CSS reflects the intrinsic geometric and shape attributes of the image. To improve the detection efficiency， we use the principal component analysis （PCA） to reduce the dimension of the HOG and CSS. The INRIA data set is used as training sample to train SVM. Compared with single use of HOG， this method has been effectively improved in detection speed and accuracy. The experimental results also verify the effectiveness of the algorithm.

Key words： pedestrian detection； HOG； CSS； fusion 1 背景

本文来源：https://www.bwwdw.com/article/9f58.html