Traffic Scene Understanding using Sound-based Localization, SVM Classification and Clustering

Abstract

The thesis is about an embedded system application aimed at identifying the semantics of traffic based on acoustic data. Sound localization, classification and clustering are used for scene understanding. The report presents a set of experiments used to simulate different traffic scenarios. An alternative implementation for sound localization is also explored, where fixed point representation of rational numbers is used instead of floating point numbers. The results for both the implementations are compared in terms of execution speed and accuracy for a Programmable System-on-Chip (PSoC).