Application of Neural Network and Kalman Filtering in Sound Source Localization

Regarding the recognition of sound beacon by the Beacon Group of the 15th National College Student Smart Car Race, the data collected by the microphone can be used to determine the sound source location through the TDOA algorithm. However, this method has limited positioning accuracy and there is obvious picket fence effect. In this paper, to improve accuracy in sound source localization and eliminate the influence of picket fence effect, we have innovatively designed a sound source localization method that first uses neural network to perform nonlinear fitting, and then uses Kalman filtering to fuse IMU data into angle calculation. This method significantly improves the accuracy in sound source localization, which has certain reference significance for the development of sound source localization methods.