基于局部行颜色自相似性特征的行人检测方法.pdfVIP

Microcomputer Applications Vol.31, No.4, 2015 研究与设计 微型电脑应用 2015 年第31 卷第4 期 文章编号：1007-757X(2015)04-0004-04 基于局部行颜色自相似性特征的行人检测方法 翟芬，蔡烜，李一蒙，冯瑞 摘 要：针对计算机视觉领域的行人检测问题，提出一种基于局部行颜色自相似性特征，该特征可表征为在HSV 空间，图 像水平方向非重叠对称块颜色直方图的距离信息，结合多层次导向边缘能量特征形成图像的融合特征，利用交叉核支持向量 机进行分类。与主流用于行人检测的HOG ＋SVM 方法相比，其特征维数低，在保证检测精度的同时，大幅提高了算法效率。 实验结果验证了该算法的有效性。 关键词：行人检测；局部行颜色自相似性；多层次导向边缘能量特征；交叉核支持向量机 中图分类号：TP391 文献标志码：A A Method of Pedestrian Detection Based on the Local Row Color Self-similarity 1 1,2 2 1 Zhai Fen , Cai Xuan , Yimeng Li , Feng Rui (1.Video Technology and Systems Engineering Research Center of Shanghai, School of Computer Science, Fudan University, Shanghai 201203, China; 2.The third Institute of Ministry of public security, Shanghai 201204, China) Abstract: For pedestrian detection problems in computer vision, this paper proposes a feature based on the local row color self-similarity. In HSV space, this feature represents the color histogram distance of the symmetric non-overlapping blocks in the horizontal direction. It combined Multi-Level Oriented Edge Energy Features with this feature to obtain fusional features, and used Histogram Intersection Kernel Support Vector Machine to classify. Compared to the method of mainstream HOG+SVM, the dimen- sion of this feature is lower. While guaranteeing the detection accuracy, the efficiency of this method is improved mostly. Experiment results validate the effectiveness of the proposed approach. Key words: Pedestrian Detection; Local Row Color Self-similarity; Multi-level Oriented Edge Energy Features; HIKSVM 0 引言 来刻画行人的局部轮廓特征，称为 “edgelet ”特征，使用 行人检测可定义为：判断输入图片（或视频帧）是否包 Boosting 来学习行人头部、躯干、腿和整个人。另一个形状