基于紧急异常声音事件检测与分类的音频监控系统方法研究

项目名称： 基于紧急异常声音事件检测与分类的音频监控系统方法研究

项目编号： No.61305027

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 倪崇嘉

作者单位： 山东财经大学

项目金额： 24万元

中文摘要： 本课题选择以商城、车站、广场等公共场所作为典型展示区域，研究基于紧急异常声音（如人的尖叫声、呼救声、爆炸声、撞车声、玻璃破碎声、自动报警声等）检测和分类的音频监控系统。1）拟研究结合听觉感知和数据驱动的特征抽取算法；2)拟研究基于目标声音事件和相似于目标声音事件的分层异常声音事件检测和分类算法；3）拟研究背景模型自适应方法；4）拟研究Novelty事件的检测和建模方法；5）拟提出基于图形处理器的异常声音事件检测和分类系统的并行实现方案。本研究的预期研究成果可为音频信息处理提供新的思路、新的分析工具和新的特征抽取手段。不仅可以解决音频检测和分类的应用难题，还可以广泛应用于公共安全领域，有助于社会的和谐稳定。

中文关键词： 音频事件检测与分类；长短时记忆递归神经网络；分层算法；背景模型自适应；特征增强

英文摘要： In this project, we select the shopping plaza, station, and square as typical display field,and study audio surveillance system based on the emergency abnormal acoustic events, such as scream sound, cry for help, explosion sound, car carsh sound, glass break sound, and personal alarm sound, detection and classification.We intend to 1) study the feature exaction algorithm combining the auditory and dada-driven; 2) to study the hierarchical abnormal acoustic event detection and classification algorithm based on the target acoustic event and similar target acoustic event; 3) to study the background model adaptation; 4) to study the novelty acoustic event detection and modeling; 5) and finally to propose the parallel implementation scheme for abnormal acoustic event detection and classification. The expected research achievements of this project provide new ideas, new analysis tools and new feature extraction means for audio information processing research. It can not only be used to solve the application problems existing in audio detection and classification, but also be widely used in public security area, and contribute to social harmony and stability.

英文关键词： Acoustic event detection and classification；Long short-term memory recurrent neural network；Hierarchical method；Adaptation of background model；Feature enhancement