Intelligent Audio And Video Processing System Based On Microphone Array

Intelligent video surveillance is one of the most popular application fields in computer vision technology.It has a wide range of applications in finance,transportation,urban security,and military fields.Because of the camera has defects such as monitoring "blind angle" and obstructing the target,some scholars put proposed that added the audio signal as supplementary signal of video signal to the monitoring system,and the speech signals played a very good supporting role.The related theoretical algorithms and practical system research and development has a high theoretical value and a wide range of applications.The thesis analyzes and compares existing microphone array sound source localization and video target monitoring methods.On this basis,we focus on the sound source localization methods based on time-delay estimation.And a video target monitoring experimental system based on the microphone array is established based on the experimental platform of moving target detection,recognition and tracking developed by the laboratory.The system is suitable for indoor scenes without a dedicated staff and it has achieved better experimental test results.The main work of this thesis includes the following aspects:(1)Describes the microphone array signal model under ideal conditions,noise and reverberation conditions.Several different time delay estimation methods are studied in depth.An improved phase power delay estimation algorithm is proposed,which can effectively remove the unreasonable delay estimation results and improve the accuracy of the delay estimation.The effectiveness of the improved algorithm is verified through simulation experiments and real environment experiments.(2)The far-field model and near-field model of array signal are described.The basic principles and implementation steps of several sound sources localization methods are introduced.The methods of angular distance location and sphere interpolation are emphatically discussed.On the basis of this experiment,the experiment of locating the moving sound source was carried out,and the microphone calibration work was added in the experiment to estimate the position of the sound source more accurately.(3)Three common video object detection methods are studied,and the principle and algorithm steps of related video processing algorithms are introduced and analyzed.Morphological operation is used to perfect the shape and contour of the target,several common features are discussed,the upper quarter feature introduced into this system is introduced,the target is trained and classified by support vector machine,has a good recognition effect.The tracking method based on mean shift in the moving target tracking method is discussed,and the edge adjustment function is added in the tracking process so that the moving target is always in the video frame.(4)Built a combination of audio and video signal monitoring system for night unattended indoor monitoring.During the operation of the system,when the sound signal is collected,the improved cross power spectral delay estimation algorithm and the angle distance positioning method are used to locate the sound signal and control the ball machine to turn to the azimuth.Then,the system processes and analyzes the video data what captured by the dome camera to determine whether there is a moving target and whether the moving target is a human,respectively,so as to judge whether an abnormal condition occurs and start the corresponding alarm to alert the security personnel.