Implementación de un microprocesador de aplicación específica para la ejecución del algoritmo de modelos ocultos de Markov en el reconocimiento de patrones acústicos.

Abstract

In this work, the classification unit of an acoustic pattern recognition system based in the algorithm of Hidden Markov Models (HMM) was developed for the detection of gunshots and chainsaws in a protected tropical area, with the idea of later building a surveillance wireless sensor network with similar nodes. The HMM was implemented with a solution that integrated hardware and software. For this reason, an application-specific microprocessor was implemented and testing in a FPGA. The recognition rate obtained was: 90,33% for chainsaw and 85,43% for gunshot. Results of the functional verification compared with to reference model are shown. Some post-place-and-route results also are presented: the device utilization summary, operating frequency and power consumption.