Study On Sound Source Localization For Mobile Robot Based On Time Delay Of Arrival

Along with the increasing expansion of application fields, mobile robots are facingincreasingly complex work environments. In all uncertain and unstructured environments whicha variety of information presented, mobile robots only rely on visual means to obtain information,is bound to reduce the environmental adaptability of the mobile robot, and thus limit their levelof intelligent decision. As the robot’s organ for processing sound information, the robot auditorysystem which is built from the idea of humanoid, has become an important research object inrobotics field. By using the physical characteristics of sound propagation, robot auditory systemcan be compatible with the visual system to compensate for its limitations, such as theperspective is limited, dependent on light and can’t pass through the non-translucent obstacles.Through researching target localization which is a basic problem in the field of environmentalperception, this thesis aims to realize mobile robot’s sound source localization based on timedelay of arrival in noise conditions by organically combining mobile robot technology withsound source localization technology. So, the following four research works have been done inthis thesis.(1) Based on the study of sound source localization method which is using microphonearray, in order to achieve accurate positioning for the sound target in3D space, and consideringthe system’s real-time, a tetrahedral microphone array is developed and produced, and then itsobtained sound information pass to mobile robot through data acquisition system, so a completerobot auditory system has been realized.(2) To solve the problem that there are many noises under actual working conditions, andimprove the performance of anti-noise and reverberation, two basic estimation methods for time delay of arrival, generalized cross correlation estimation and adaptive filtering, are studied. Andthe paper analyzs the performance of typical methods through simulation experiments,summarizs the application occasions, the advantages and disadvantages of the typical algorithms.Then the time delay of arrival estimation method based on the combination of generalized crosscorrelation and adaptive filter is studied.(3) According to the tetrahedral microphone array structure and sound propagation sphericalwave model, precise geometric localization calculation model has been established. On this basis,the far-field sound source localization calculation model can be derived. After performancesimulation analysis, a full-ranged sound source orientation method combined with robot’s activemovement has been presented. And by using neural network’s capability of nonlinear curvefitting and parallel computing, a sound source orientation calculation model based on BP neuralnetwork has been also presented.(4) To solve the problem which a large number of outliers exist caused by sensorobservation uncertainty and sudden strong interference, considering the relative motion betweenmobile robot and sound source, an anti-outliers double Kalman filtering method based oninnovation has been presented. Experimental results show that this method can improve system’sreliability by adjusting measurement noise covariance during continuous sound source targetlocalization and tracking.