| With the deepening of the aging trend of our country’s population,the safety monitoring of the elderly living alone has become a research hotspot.In this paper,the method of combining fall sound recognition and sound source localization technology is employed to detect the fall behavior of the elderly.The recognition of fall sound by Mel Frequency Cepstrum Coefficient(MFCC)and location of fall sound are mainly studied.Firstly,the classification and recognition of fall sounds are studied.In order to reduce the amount of calculation,a dual-threshold endpoint detection method combining short-time energy and short-time zero crossing rate is applied to detect silent and vocal segments,and only the latter is processed.The preprocessing of the voiced segment is made to compensate for the attenuation of the high-frequency part during the transmission process and to satisfy the smooth signal required for subsequent processing.Preprocessing includes pre-emphasis,framing,windowing.After preprocessing,the 39-dimensional MFCC in each frame is extracted to represent the fall sound feature,which can be used as training of naive Bayes classifier or as recognition of fall sound.Secondly,the pitch angle and azimuth angle of the sound source need to be calculated by the sound source localization algorithm to further determine whether the identified fall sound is emitted when a person falls.The search space is divided into grids,and the power of all nodes in the spatial grid is calculated by beamforming to form the spatial spectrum.The grid position corresponding to the maximum value in the spatial spectrum is the estimation of the sound source position.The disadvantage of grid search method is that the higher the search accuracy,the more power the nodes stored,the greater the memory consumption,and the longer the search time.To solve the above problems,the sparrow search algorithm is used in the spatial search process in this paper.By randomly searching the space and updating and recording the optimal value of power spectrum in each iteration,the effect of dimension reduction is achieved,which directly reduces the memory consumption and the increase of calculation time caused by the increase of search accuracy.Finally,the accuracy of falling sound recognition and the deviation of sound source localization were tested in the actual scene,and then multiple sets of experiments were carried out by combining the sound recognition module and the sound source localization module to verify the accuracy of the system for fall behavior detection. |
