Embedded realtime feature fusion based on ANN, SVM and NBC
Andreas Starzacher, Bernhard Rinner · 2009
Abstract – Artificial neural networks (ANNs), support vector machines (SVMs) and naive Bayes classifiers (NBCs) are common tools for multisensor data fusion applications. In this paper ANN, SVM and NBC are applied to embed-ded realtime feature fusion and compared to different algo-rithms concerning classification execution time as well as classification rate. These algorithms are implemented on our three-layered multisensor data fusion architecture and applied to traffic monitoring where we are focusing on fus-ing data originating from distributed acoustic, image and laser sensors for vehicle classification and tracking. The evaluation of the algorithms is performed on our em-bedded platform and has shown promising results concern-ing realtime classification execution time and classification rate.