基于视觉注意机制的行人目标检测  

The Research of Pedestrian Detection Algorithm Based on Visual Attention

作　　者：赵谦[1] 薛改样 杨新花 ZHAO Qian;XUE Gai-yang;YANG Xin-hua(School of Communication and Information Engineering,Xi'an University of Science and Technology,Xi'an Shanxi 710054,China;Shenzhen Kelie Technology Co.,Ltd. Shenzhen Guangdong 518057,China)

机构地区：[1]西安科技大学通信与信息工程学院,陕西西安710054 [2]深圳市科列技术股份有限公司,广东深圳518057

出　　处：《计算机仿真》2019年第7期411-414,共4页Computer Simulation

基　　金：陕西省科技计划工业科技攻关(2017GY-073,2015GY-023);西安市碑林区应用技术研发(GX1811)

摘　　要：在分析现有运动目标检测算法的基础上,针对三帧差分检测易出现空洞及边缘不连续等问题,提出一种基于视觉注意机制的行人目标检测算法。该算法首先,通过借鉴人类视觉显著性现有的研究成果获得视觉显著图;其次,使用一种逻辑迭代三帧差分法,将其与基于区域搜索迭代的内轮廓填充法相结合,获得运动区域,融合两种算法提取出ROI;最后,采用HOG特征结合SVM分类器对ROI进行行人检测。对比实验结果表明,该算法连通性好,准确率高达97%以上。Based on the analysis of the existing moving target detection algorithm, for the phenomena in three - frame difference detection, such as cavity and edge discontinuity ,, a pedestrian object detection algorithm based on visual attention mechanism was proposed. Firstly, we used the existing research results of human visual saliency to obtain the visual saliency map;Secondly, a logical iterative three-frame difference method was used to obtain the motion region combine with a contour filling method based on the local region research, two algorithms were integrated to obtain the region of interest. Finally, HOG features was used combined with support vector machines classifier for pedestrian detection in region of interest. The experimental results show that the algorithm has good connectivity and the accuracy is over 97 %.

关 键 词：视觉显著图 三帧差分法 轮廓填充法 感兴趣区域 支持向量机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]