摘要
采用基于人体HOG特征提取算法,对变电站安全管控系统状态监测视觉辨识技术进行研究。根据变电站具体环境,人体特征等现象,采用级联Adaboost分类器,经离线训练和在线分类,可准确、快速实现对变电站状态监测视觉辨识,从而提高系统的技术性能,使系统具有较强的实用性。实验结果表明,采用的状态监测视觉辨识技术人体检测算法检测率为93.8%,误检率为4.7%,平均耗时为62 ms,比SVM分类器的检测率要高出9.5%,误检率要低9.8%,平均耗时要少132 ms。采用级联Adaboost分类器检测性能得到提高,从视频序列中能快速、准确提取人体区域,较好地满足了动态目标检测、分析的需求。
An HOG feature extraction algorithm based on human body is used to study the state monitoring and vision identification technology for the safety management and control system of substation. According to the specific environment of substation,human characteristics and other phenomena,the substation state monitoring and vision identification can be achieved rapidly and accurately by means of online classification and offline training of the cascade Adaboost classifier,so as to improve the system technology performance,and make the system practicability stronger. The experimental results show that the detection accuracy of the human detection algorithm based on state monitoring and vision identification technology is 93.8%,its false detection rate is 4.7%,and its average consuming time is 62 ms. In comparison with SVM classifier,its detection accuracy is 9.5%higher,the false detection rate is 9.8% lower,and the average consuming time is 132 ms shorter. With the cascade Adaboost classifier,the detection performance can be improved,and the human body region can be extracted in the video sequence quickly and accurately,which can meet the requirements of dynamic target detection and analysis.
出处
《现代电子技术》
北大核心
2017年第11期80-83,共4页
Modern Electronics Technique
