Development and evaluation of a scheme for detecting multiple approaching vehicles through acoustic sensing

We propose a robust scheme for detecting vehicles approaching from several directions to prevent crossing collisions. The scheme consists of three sequential operations. The first operation computes cross-power spectrum phase (CSP) coefficients of sound signals from a microphone array; the second operation classifies the information obtained by the CSP coefficients into multiple ranges of directions so as to separate multiple approaching vehicles. The last one detects approaching vehicles through a threshold operation of the clustered information. In order to evaluate the performance of the scheme, we constructed a large-scale database based on the acoustic and related information concerned with approaching vehicles. Using the database, we show that the scheme is robust for detecting not only a single but also multiple approaching vehicles.