The Design And Implementation Of Sound Source Localization System Based On Small Microphone Array

In recent years,with the development of artificial intelligence,the intelligentization of robots has been further advanced,and more service robots have been involved in people's life.In order to enable robots to deal with complex work environments better,they need to acquire real-time environmental information and handle it in time,while the sound is the most important information to get and positioning the sound source is a key algorithm for robot to response.Based on existing robotics projects in the laboratory,the author designs a sound source localization system according to a small microphone array to solve the problem existing in robot positioning of sound sources.In this thesis,the author introduces three kinds of classical microphone array-based sound source localization(SSL)methods,analyzes their principles,applicable fields,advantages and disadvantages,and finally selects Steered Response Power with Phase Transform(SRP-PHAT)method which combines the stearing beam formation technology and phase transform as a referenced implementation plan.Through experiments,it has been found that the SRP-PHAT method has a long positioning time and does not meet the need for real-time positioning of the robot platform,and the problem lies mainly in the search method of sound source in the SRP-PHAT.This thesis puts forward the Particle Swarm Optimization(PSO)to improve the traditional search method of sound source and proves that the PSO is more effective for searching sound source by simulation experiments.In the end,the author designs a microphone array and constructs a basic hardware environment based on xCore-200 platform.Based on this,the traditional search method and the improved search method are implemented,and experiments are performed to analyze azimuth accuracy,distance accuracy and real-time for the two methods in the horizontal two-dimensional.As experimental results show,the SSL system implemented on xCore-200 behaves that most success rates of azimuth estimations can be over 90%,and the average error can be controlled within 5°;The distance estimate is slightly worse,with an average estimated success rate near 80% and an average error around 0.25 meters;in terms of time cost,the average single SSL time is 525.23 ms.Compared with traditional grid search method,the optimized method keeps the same accuracies of azimuth and distance estimation,and the average SSL time is shortened by 20.48%.