用于安全监控系统的音频事件检测研究
2017-05-30夏亦佳
夏亦佳
摘 要:针对视频监控过于依赖图像信息,在黑暗条件下或视线盲区无法进行及时、有效监控的问题,本文提出一种可以用于辅助安全监控的音频事件检测系统,以枪击声和尖叫声为关键事件,把过零率、短时能量、子带能量比和Mel倒谱系数作为音频特征,基于SVM分类器设计实现了一种多级分类系统。结果表明,该系统可以有效地检测出两种关键事件,最优识别率达90%。
關键词:音频事件检测;安全监控视频;支持向量机;MFCC
中图分类号:TN912 文献标识码:A 文章编号:1003-5168(2017)06-0015-04
Audio Events Detection for Surveillance System
Xia Yijia
(China Airborne Missile Academy,Luoyang Henan 471009)
Abstract: Public surveillance system relies on image information to a large extent, therefore, places in dark environments or blind areas of surveillance cameras would not get effective real-time surveillance. To solve this problem, this paper proposed an audio events detection system to assist video-based public safety surveillance. The system took the shooting and screaming as the key events, and defined Zero Crossing Rate (ZCR), Short Time Energy (STE), Sub Band Energy Ratio (SBER) and MFCC as audio features. Then a hierarchical SVM-based classification system is implemented to classify different types of audio events. According to experiment results,the proposed audio events detection system can effectively classify gunshots and screaming sound from environmental noise with an optimal classification accuracy of 90%.
Keywords: audio events detection;surveillance system;support vector machine;MFCC
随着多媒体技术的飞速发展,越来越多公共场所安装了视频监控以保障公共安全。然而,在黑暗环境下或者在视频监控盲区,这些依赖于图像信息的摄像头就难以进行及时、有效的监控。此时,通过音频传感器的监控则会更加有效[1]。声音信息中也包含了不亚于图像信息的信息量,通过监控识别公共场合声音信息,可以迅速发现所关注的事件,从而快速做出应对。声音信息具有良好的实时性,对视频监控具有良好的辅助作用。
本文提出了一种基于支持向量机(Support Vector Machine,SVM)的音频事件检测系统,选取枪击声和尖叫声为关键音频事件,提取训练样本的过零率、短时能量、子带能量比、MFCC组成特征向量对SVM进行训练,建立一个音频事件分类模型,对测试样本中的关键事件进行检测。
1 音频事件检测系统
音频事件检测系统如图1所示。音频事件检测主要分为声音信息的特征值提取、参数学习和关键事件检测三部分。……
