Multiple Sound Source Localization Method And Its Application In Vehicle Honking Monitoring Systems

With the increasing development of array signal processing technology,the sound source localization(SSL)technology based on microphone array has been applied in many industries such as human-robot interaction,mechanical fault detection,audio and video conferences,and medical equipment.More recently,the SSL technology is also being used to monitor vehicle honking under urban road traffic circumstances.At the present stage,the research of horn honking monitoring system mainly focuses on single source localization problem,which fails when dealing with multiple-vehicle honking.Therefore,the main research task of this thesis is to improve the localization accuracy of multiple horn sound by studying and improving the SSL algorithm.With the help of the environment sound recognition(ESC)technology,we also eliminate the interference of other environment sounds to the location result.The main works of this thesis are as follows:Firstly,the ESC technology based on artificial neural network is studied.The signal preprocessing method,the feature extraction method of Mel-frequency cepstral coefficient(MFCC),and the structure and learning algorithm of back propagation neural network are introduced and derived in detail.The MFCC features extracted from the environmental sound signal are input to the established three-layered BP neural network for training,and the results show that the recognition accuracy of this method is above 96%.Secondly,the design method of the microphone array and the traditional SSL technology are studied.Based on the microphone array with a uniform circular structure,the principle and mathematical model of the delay and sum beamforming(DSB)localization algorithm are introduced,and the factors that influence the localization performance are analyzed and verified by simulation from two aspects of beamwidth and spatial sampling theorem.Considering the frequency domain characteristics of vehicle horn sound signal and the factors affecting the localization accuracy,a 16-element planar microphone array with a uniform concentric circle geometry is designed,which can reduce the influence of the array aperture on the localization performance.Thirdly,based on the DSB algorithm,this thesis discussed and studied the multiple sound source localization problem,and then proposed two improvements for the purpose of improving the angular resolution and suppressing the maximum peak side lobe(PSL).First of all,based on the principle that the side lobe of high frequency signal is larger in large aperture array and the beamwidth of low frequency signal is wider in small aperture array,an improved method named frequency classification processing is proposed.The frequencies with large peak are extracted from the received signal as the location frequencies,which are divided into higher frequency components and lower frequency components according to the preset threshold value.Then based on the DSB algorithm,for the inner circular array,use the higher frequency components of the received signal to calculate its beam output,similarly,the beam output of the outer circular array is determined by the lower frequency components of its received signal.The second strategy is as follows: the particle swarm optimization(PSO)algorithm is applied to the adaptive weighted processing of the beam output of each location frequency to further reduce the beam width,which can meet the requirements of multi-sound source localization accuracy.Based on practical application background,parameters such as the main lobe width,peak main lobe and PSL in the beam pattern are selected to construct the fitness function of PSO algorithm,and iteratively updates the weights by tracking the direction of decreasing fitness value.Lastly,according to the location relationship between the horn honking monitoring system and the vehicles under the actual urban road environment,a scaled down experimental platform is built in a semi-anechoic chamber.Multiple sound source localization experiments and minimum angular resolution tests are performed with different methods,i.e.,traditional beamforming localization algorithm based on a uniform circular array,and improved beamforming algorithms based on a uniform concentric circular array.The experimental results show that when the angle between two adjacent sound sources is reduced to 6°,the improved algorithm can still accurately estimate the direction of two sound sources,and the average estimation error is within 2°.