Neural Network Approach to Acoustic Detection of Number of Vehicles

摘要

The paper reports results of applying a perceptron neural network to determine the number of sound sources on a scene monitored by an array of linearly spaced microphones. The standard techniques for solving this problem were found inadequate in the presence of normal disturbances (such as produced by wind). The paper proposes an indirect application of a perceptron neural network, to analyze the results of the MUSIC beam forming technique. The method is experimentally shown to deal with this problem. Field experiments included scenes with zero, one or two moving vehicles proved the system effectiveness.