Research On Single Sound Source Localization And Tracking Technology Based On Sound Energy

The source localization and tracking is one of the fundamental works for acoustic event perception and understanding,which has gained attention and applications in both breadth and depth,especially in industrial fault detection,intelligent device interactio n,and underwater target monitoring.For its directness and effectiveness,many researchers are increasingly exploring techniques related to acoustic energy-based source localization and tracking,and have produced many valuable works.Thanks to these results,an indepth theoretical study on the estimation of the position and moving trajectory of a n acoustic source has been presented in this paper from the perspective of combining theory and engineering.The main research work in this paper has the following four aspects.First,this paper has designed a specific calibration scheme to calibrate the propagation attenuation model of acoustic energy.With the introduction of the attenuation coefficient into the traditional propagation model,an objective function can be established for its calibration based on the least-squares criterion,using the acoustic energy from different collecting points.According to the calibration scheme,the attenuation coefficients of different collecting points in the Hemi-anechoic chamber were estimated and smoothed by numerical averaging to obtain a reliable estimate of the attenuation coefficient.In addition,the experimental results in the Hemi-anechoic chamber also showed that 1)the attenuation coefficient is smaller in the positive direction of the acoustic source propagation;2)the farther the collecting point is from the source position,the larger the attenuation coefficient is.Second,based on the calibrated energy propagation attenuation model,t his paper has built a theoretical analysis framework for single-source localization in general scenarios using Bayesian estimation.In this framework,the placement deviations of acoustic sensors and the layout of acoustic sensing network arrays(effective number of acoustic sensors,array structure)have been introduced,in addition to expanding the randomness of ambient noise in time and space.Further,the asymptotic effects of these common influencing factors on the localization performance of a single acoustic source are revealed by the analysis of the Cramer-Rao lower bound.In addition,the Cramer-Rao lower bound was also provide theoretical guidance on the layout of the acoustic sensing network array.The simulation results showed that the uniform ar c array(UAA)and grid array(GA)had better localization performance among the regular centralized and distributed arrays,respectively.Overall,the GA has the best performance and has the "flat bottom phenomenon" in a coverage area.Based on the above generalized theoretical analysis framework,this paper further characterized the sound source and noise correlation using a fractal Gaussian noise(f Gn)statistical model to investigate the realization of single-source localization in a correlated noise environment.Due to the introduction of common influencing factors,the posterior probability density function of the acoustic source position is non-convex.A particleassisted random search algorithm was used to achieve an effective estimation of the single-source position.In addition,this paper computed the non-stationarity degree of the acoustic source and interference noise using the index of non-stationarity(INS).Using acoustic sensing network GA for localization,two sets of experiments showe d that: 1)the influence of common influencing factors on single-source localization in real scenes is consistent with the theoretical analysis results in the previous work;2)the f Gn statistical model is more effective in portraying the correlated noise environment than the traditional Gaussian white noise model;3)the higher the degree of non-smoothness of the acoustic source,the worse the localization accuracy of the single source.Finally,again using the calibrated energy propagation attenuation mod el and dynamically extending the single-source localization framework,this paper has developed a study of single-source tracking.The movement of the acoustic source can be categorized as regular movement(e.g.,uniform linear motion)or random movement(e.g.,Brownian motion).The process of the two types of movement can be exactly described as the acoustic source position model,the acoustic source position transfer model,the transfer velocity model,and the transfer deviation model.Further,using the observed energy of the acoustic sensing network array,a multilayer Bayesian network framework for single-source tracking was built based on Bayesian estimation theory.By variational inference,the moving state and moving velocity of the sound source have been estimated from this framework.In addition,the performance limit of single-source tracking was also derived in this paper using the Cramer-Rao lower bound.An acoustic sensing network GA was used,and the simulation results showed that the multilayer Bayesian network framework and the variational inference method can provide effective tracking of the two types of moving processes.In addition,the effect of noise and the number of acoustic sensors on the tracking results were also analyzed,and the "effective coverage phenomenon" regarding the acoustic sensing network array was observed.