Abstract
For the purpose of developing acoustical environment recognition system in autonomous mobile robots, a practical method for simultaneous estimation of 3D-position and velocity of a moving random acoustic source in near-field is proposed by analyses of wavelet patterns of signals detected at 4 fixed points. By analysing correlation of time-scale (time-frequency) 2D-patterns of wavelet transforms of detected signals, information on ratios of relative compression and/or extension rates of the patterns due to the Doppler effect, those of the amplitude attenuation of wave propagation, and differences of wave propagation time are derived independently to the shape of power spectrum of the source radiating random signal. As a result, it can be applied to any moving acoustic source, radiating a stationary random signal. Then, with an aid of geometric relations of detection, the source position and velocity are estimated by the least mean squared (LMS) method on the basis of the information extracted above. After the principle of the proposed estimation method and the required conditions being made clear, the effectiveness and special features of the proposed method are ascertained through the fundamental experiments.
