摘要
针对高准确率和快速跟踪摄像功能的需求,基于机器视觉设计了人形识别安防摄像机。首先,确定了整体工作方案;其次,提出一种多特征组合人形识别方法,通过训练3个分类器,分别识别人形正身、侧身和背身,并采用haar特征+adaboost算法实现人形识别和定位功能,再利用PID算法控制步进电机和舵机实现人形跟踪;最后,开发了安卓手机端APP。测试结果表明多特征组合识别方法能够快速识别人形目标,帧率不低于20Hz;与单特征人形识别方法相比,识别准确率提高27.3%;人形目标追踪时间不超过1.7s;摄像机能够流畅实现WiFi控制、无线监视和录像下载等功能。
Aiming at the demand problem of highly accurate human profile recognition and rapid tracking,the human profile recognizable surveillance camera is designed based on machine vision.Firstly,the overall work program structure is designed.Secondly,a multi-feature detection method is proposed.Three classifiers are trained,which can classify human profile of the front,side and back.The algorithm of haar feature and adaboost is integrated,which can realize functions of human profile recognition and positioning.The PID algorithm is carried out for the function of human profile tracking with controlling the stepper motor and steering gear.Finally,the Android APP is developed.The test result shows that the method of multi-feature detection can recognize human profile rapidly with frame rate not less than 20Hz.And the recognition accuracy can improve 27.3%in contrast with the method of single feature detection.The tracking time of human profile does not exceed 1.7s.The camera can work smoothly with functions of WiFi control,wireless surveillance monitoring and video downloading.
作者
周利杰
郝瑞林
蔡国庆
孙迎建
刘辉
ZHOU Li-jie;HAO Rui-lin;CAI Guo-qing;SUN Ying-jian;LIU Hui(Hebei University of Water Resources and Electric Engineering, 061001, Cangzhou, Hebei, China;Hebei Technology Innovation Center of Industrial Manipulator and Reliability, 061001, Cangzhou, Hebei, China)
出处
《河北水利电力学院学报》
2021年第4期66-71,78,共7页
Journal of Hebei University Of Water Resources And Electric Engineering
基金
河北省高等学校科学技术研究青年基金项目(QN2021226)
省属高等学校基本科研业务费研究项目(SYKY2009)
河北省大学生创新训练项目(S201910085009,S201910085010)。
关键词
机器视觉
人形识别
人形跟踪
HAAR特征
ADABOOST
machine vision
human profile recognition
human profile tracking
haar feature
adaboost algorithm
