智能监控前端系统中异常声音检测的实现  

Implementation of Abnormal Sound Detection in Intelligent Surveillance Front-end System

作　　者：张璐璐[1] 陈耀武[1] 蒋荣欣[1] 

机构地区：[1]浙江大学数字技术及仪器研究所,杭州310027

出　　处：《计算机工程》2014年第1期218-221,227,共5页Computer Engineering

基　　金：国家"863"计划基金资助项目(2010AA09Z104)

摘　　要：针对智能监控前端系统中异常声音检测的高实时性和高准确率要求,提出一种基于混合特征参数和改进动态时间弯折(DTW)算法的异常声音检测方案。通过短时幅度和过动态门限率判决声音端点,提取包括短时幅度、美尔倒谱系数和差分系数在内的混合特征参数,采用改进的DTW算法进行声音识别。在TI TMS320DM368处理器平台上的实验结果表明,基于该方案的智能监控前端系统对异常声音的识别时间小于1 s,准确率达到89.3%。Aiming at the requirements of high real-time and high accuracy for abnormal sound detection in intelligent surveillance front-end system, this paper presents a scheme of abnormal sounds detection based on mixed characteristic parameters and improved Dynamic Time Warping（DTW） algorithm. This system detects endpoints of sounds based on short-time magnitude and short-time threshold-crossing rate, extracts mixed characteristic parameters including short-time magnitude, Mel Frequency Cestrum Coefficient （MFCC） and difference coefficient. It recognizes sounds by improved DTW algorithm. Experimental results on the TI TMS320DM368 processor platform show that the recognition time of intelligent surveillance front-end system based on the proposed scheme is less than 1 s and average recognition rate is 89.3%.

关 键 词：前端系统 异常声音 实时性 混合特征参数 动态时间弯折 智能监控 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]