| The security of public places has always been the focus of people’s attention. Under normal circumstances, a surveillance video system can be a very good identification of security risk events, but the video surveillance areas often have blind spots, and in the case of view obstruction or other adverse conditions, such as rain, fog or smoke, the video surveillance can not accurately detect the abnormal events. The new generation of monitoring equipment usually has the function of video, audio and other signal acquisition, but the spread of the audio signal is not restricted by space. Moreover as general risk events are accompanied with certain abnormal sound, we can capture the abnormal feature of audio signal, combined with the functions of remote control angle rotation, zoom and so on which monitoring platform has, the use of audio signal detection and localization of early warning can be the supplementary of video surveillance and form a more effective collaborative analysis system. As the sub-topic of the "863 Project" commissioned by a Research Institute of the Beijing CASIC Second Institute, this research will provide a necessary theoretical basis for the implementation of the intelligent monitoring system through studying abnormal sound source detection and localization.In this paper, firstly, from the time domain and frequency domain analysis of audio, we analyzed the characteristic parameters of the various possible abnormal sounds in real life, and established the abnormal sound database to detect and classify the abnormal sound. Based on the improved endpoint detection algorithm, the algorithm of short-time energy time threshold is proposed, and the numerical experiments and analysis are carried out.Secondly, considering large-scale monitoring system with a huge amount of audio and video signal processing should be real-time, we discussed the parallel processing means, and improved the detection part of the serial algorithm; Subsequently, the parallel program designed by using Studio Visual 2010 based on OpenMP library on the quad core CPU was given. At last, a method is offered as to choose suitable microphones to determine sound source location.As the sound source localization research of current microphone array was in the fixed small positioning system, the microphones of the monitoring system studied here had large distances, and the four element array of localization microphone is not fixed. This article used the time delay estimation of the sound source localization method to establish the closed-form linear localization model, and gives the existence and uniqueness of the solution, the effectiveness of which is again proved by the ensuing numerical experiment results and analysis.Finally, based on the results obtained from a series of discussions, we designed and developed the abnormal point sound source detection and positioning system in the monitoring system based on Visual Studio 2010 with MFC interface. |
