Design of a vehicle telemetry system validated by AI-driven driver behavior analysis

Abstract

Most road accidents are caused by human error, so detecting unsafe driving in real time has become a key road-safety goal. This work, carried out with the Centre de Développement des Technologies Avancées (CDTA), presents an embedded platform that acquires, records, and analyses driving data on board the vehicle. The device reads signals from the CAN/OBD-II interface, an Inertial Measurement Unit (IMU), and a GPS receiver, stores them on a microSD card, and on the computer side where an offline Artificial Intelligence (AI) model classifies driving as safe or unsafe based on friction circle. Three machine-learning models — Support Vector Machine (SVM), Random Forest, and XGBoost —were evaluated and validated. On the software side, a FreeRTOS-based firmware manages the several tasks of acquisition, logging, and real-time streaming, while the custom Printed Circuit Board (PCB), built around the ESP32-S3 microcontroller, ensures reliable communication with all subsystems.