基于音视频分析的区域安防管控平台
2019-10-21蔡烜蒋龙泉冯瑞
蔡烜 蒋龙泉 冯瑞



摘 要: 针对广场等区域人流量大、安保要求高的特点,设计并实现了一套基于音视频识别的区域安防管控平台。该平台从区域安防的需求出发,设计了嫌疑人员报警、人群拥挤报警和异常声音报警三大功能,以人脸识别、人群密度估计、异常声音识别三个音视频识别算法作为支撑,实现对区域内与人相关的异常事件的预警。首先介绍了基于深度神经网络的人脸识别算法,然后介绍了一种引入注意力机制的卷积神经网络模型来实现的人群密度估计算法,和一种基于多卷积神经网络模型融合的异常声音识别算法;最后介绍了平台需求与设计过程,主要包括平台建设的需求分析和界面设计。
关键词: 视频分析;音频分类;深度学习;人脸识别;人群密度估计;异常声音识别;区域安防管控;
中图分类号: TP311
文献标志碼: A
文章编号:1007-757X(2019)06-0017-04
Abstract: We design and complete a regional security control platform based on audio and video recognition in order to protect people in the area from dangerous situation. The platform is designed three functions: suspect alarm, crowd crowding alarm and abnormal voice alarm. It uses face recognition, crowd density estimation and abnormal voice recognition as the support to realize the early warning of abnormal events in the region. Firstly, it introduces the face recognition algorithm based on deep neural network, then a convolutional neural network model with attention mechanism is introduced to realize the population density estimation algorithm. Finally, it introduces an abnormal voice recognition algorithm based on multi-convolutional neural network model fusion. At last it introduces the platform requirements and design process, including the platform construction requirement analysis and interface design.
Key words: Video analysis; Audio classification; Deep learning; Crowd density estimation; Abnormal voice recognition; Regional security management
0 引言
近年来,我国经济保持高速发展,城镇化率不断提高,城镇人口增多,广场、景区、高铁站等人流量大的区域成为城市安全的重点管控区域,这些区域人员流动性大,人员身份难确认,存在各种安全隐患和威胁,现有传统视频监控手段还比较落后,监控中心无法实现实时视频全覆盖,单纯依靠人力无法及时发现区域内的风险和隐患,只能通过视频回放进行事后追查已经不符合新时期安防工作的需要。
与此同时,科技领域中音视频识别技术不断提升,如人脸识别技术、音频识别技术等,这些技术已逐步在各行各业中开始应用,在推动科技进步和经济发展同时,为安防管控新增了强有力的技术手段,区域安防管控平台也可以利用这些新技术手段提升管理水平。……
