A method of real-time mobile vehicle identification by means of acoustic noise analysis implemented on an embedded device

Abstract

The paper considers a novel method of mobile vehicle identification based on acoustic signal analysis and the implementation of the method on a specific embedded device. The algorithm is designed as a multistage decisionmaking scheme, which involves frequency domain feature extraction, fuzzy classification, correlation analysis and signal dynamics monitoring. The implementation of the system is tested in real-time conditions on an embedded platform. The results of processing time measurements indicate the ability to operate in real-time with several different signal frame lengths.

A Multistage Procedure of Mobile Vehicle Acoustic Identification for Single-Sensor Embedded Device

“…The test signal was acquired prior to the experiment. After training the fuzzy rule-base and tuning all the parameters of the algorithm, the signal file is streamed to the device input buffer bypassing the ADC at the rate of the sampling frequency in order to simulate real-time data acquisition and operation [16]. The frame length is chosen, as in the previous sections, to be 2 14 = 16384, which corresponds to 0.3715 seconds at the sampling rate 44.1 kHz.…”

“…|, the first two of which are also present in the correlation calculating equation (16). Computing these sums only once and minimizing the number of cycles during feature extraction greatly reduces the number of overall operations.…”

Object detection for military surveillance using distributed multimodal smart sensors

“…A previously developed multistage algorithm for vehicle identification [7] is applied in acoustic UGS. In [8] we show the ability of the method to operate in real-time on embedded hardware with various frame lengths. For this work we, however, use separate UGS for acoustic identification in order not to interfere with localization.…”