Authors:
J. M. Lozano Domínguez
1
;
J. M. Corralejo Mora
1
;
I. J. Fernández de Viana González
2
;
T. J. Mateo Sanguino
1
and
M. J. Redondo González
1
Affiliations:
1
Department of Electronic Engineering, Computer Systems and Automatics, University of Huelva, Av. de las Artes s/n, 21007 Huelva, Spain
;
2
Department of Information Technologies, University of Huelva, Av. de las Artes s/n, 21007 Huelva, Spain
Keyword(s):
Embedded System, Machine Learning, Performance Analysis, Road Signalling, Target Detection.
Abstract:
Embedded systems with low computing resources for artificial intelligence are being a key piece for the deployment of the Internet of Things in different areas as energy efficiency, agriculture or water monitoring, amid others. This paper carries out a study of the computational performance of a smart road detection and signalling system. To this end, the implementation methodology from Matlab® to C++ of a one-class SVM classifier with two pattern analysis strategies based on RADAR signals and RAW data is described. As a result, we found a balance between AUC, RAM consumption, processing time and power consumption for a Teensy 4.1 microcontroller with STFT and the fitcsvm2 algorithm versus other hardware options such as an I7-3770K processor, Raspberry Pi Zero and Teensy 3.6.