基于快速级联分类器的行人检测方法研究
2016-01-08高嵩,杜晴岚,陈超波
基于快速级联分类器的行人检测方法研究*
高嵩,杜晴岚,陈超波
(西安工业大学电子信息工程学院,陕西 西安 710021)
摘要:行人检测是图像处理、计算机视觉等方面研究的重要环节,通常用于视频监控和智能车辆等领域。行人检测图像易受到背景的影响,常用的帧差法及单纯训练分类器法在行人检测中存在着准确率低、分类训练算法复杂、实时性差等问题。首先采用改进型帧差法获取行人运动信息,然后利用直方图坐标对应划分出运动区域,最后通过训练双特征级联分类器对运动区域进行检测识别。实验结果表明,本方法可以有效减少误检和漏检现象,检测时间平均减少了32.77 ms,检测准确率平均提高了10%以上,因此本方法有效提高了识别准确率和识别速度。
关键词:改进型帧差法;Haar特征;Shapelet特征;级联分类器
中图分类号:TP391.4 文献标志码:A
doi:10.3969/j.issn.1007-130X.2015.06.022
收稿日期:*2014-05-23;修回日期:2014-06-30
基金项目:国家自然科学基金资助项目(61271362);陕西省国际科技合作重点项目(2015KW-024);西安市技术转移促进工程项目(CXY1441(3))
作者简介:
通信地址:710021 陕西省西安市未央区草滩镇西安工业大学电子信息工程学院
Address:College of Electronic and Information Engineering,Xi’an Technological University,Caotan Town,Weiyang District,Xi’an 710021,Shaanxi,P.R.China
Apedestriandetectionmethodbasedonrapidcascadeclassifier
GAOSong,DUQing-lan,CHENChao-bo
(CollegeofElectronicandInformationEngineering,Xi’anTechnologicalUniversity,Xi’an710021,China)
Abstract:Pedestrian detection is an important part of image processing and computer vision research, and it is usually used in the field of video surveillance and smart vehicles.Pedestrian images are vulnerable to the background,and using common methods such as frame difference and training classifier to detect pedestrian has problems such as low accuracy rate,algorithm complexity and poor real-time pedestrian detection.In this paper we use an improved frame-difference method to obtain the pedestrian movement information.Then we utilize histogram coordinates to divide the movement region correspondingly.Finally,the motion area is detected and recognized by training a double feature cascade classifier. Experimental results show that our method can effectively reduce false and missed phenomenon,the average detection time is reduced by 32.77 ms,and the average detection accuracy is increased by more than 10%.The recognition accuracy and speed are improved effectively.
Keywords:improvedframedifference;Haarfeature;Shapeletfeature;cascadeclassifier
1引言
静态背景下的行人检测通常采用背景减除、帧差法以及光流法等方法。研究人员在前人研究的基础上不断改进,提出了一些新的方法。AbdelkadirMM等人[1]提出了一种基于子空间的背景建模方法,使背景模型适应更多场景下的行人检测。YinHong-peng[2]利用直接帧差法结合三帧差法,克服单纯使用帧差法的漏检情况。XiongJing-yi等人[3]通过引入扩展相位相关技术对初始运动进行预测,提高了光流法在大规模运动和强大的噪声影响的图像中的检测率。……
