Real-Time Vehicle Roll Angle Estimation Based on Neural Networks in IoT Low-Cost Devices

Guzmán, Javier García; González, Lisardo Prieto; Redondo, Jonatan Pajares; Martínez, Mat Max Montalvo; Boada, María Jesús López · 2018 · OpenAlex-citations

DOI: 10.3390/s18072188

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the critical need for accurate, real-time vehicle roll angle estimation to support Roll Stability Control (RSC) systems, particularly for heavy vehicles prone to rollover. While RSC systems require precise dynamic data, directly measuring roll angle with low-cost sensors is not feasible. Although previous research suggests that Artificial Neural Networks (ANNs) can effectively estimate this variable, there was a lack of experimental evidence confirming whether low-cost Internet of Things (IoT) devices could perform these ANN-based estimations under hard real-time constraints (specifically ≥50 Hz) with sufficient accuracy in real-world driving conditions. The authors aimed to design an IoT architecture integrating ANNs into low-cost hardware and assess its performance against professional-grade equipment. The researchers developed an experimental testbed installed in a Mercedes-Benz van, comprising three distinct kits: a high-end Racelogic VBOX system serving as the ground truth, and two low-cost IoT kits based on a Raspberry Pi 3 Model B and an Intel Edison System-on-Chip. Each low-cost kit included an Inertial Measurement Unit (IMU) to capture lateral acceleration, longitudinal acceleration, yaw rate, and roll rate. These inputs fed into a three-layer ANN (input, hidden with 15 neurons, and output) implemented in C++ to estimate the roll angle. The software architecture utilized WiFi for communication and Network Time Protocol (NTP) for synchronization, ensuring coherent data collection across all devices. The system was designed to process data at a 50 Hz sampling rate, adhering to the strict timing requirements of vehicle safety systems. The results demonstrated that both the Raspberry Pi 3 and Intel Edison kits successfully met the hard real-time processing constraints, maintaining the required 50 Hz frequency without significant latency. The ANN-based estimations provided by these low-cost devices were highly approximated to the real values recorded by the VBOX ground truth sensor. The study confirmed that the computing capabilities of these small, low-cost single-board computers are sufficient to run complex neural network estimators reliably. Furthermore, the architecture proved robust under high dynamic conditions, with the estimators remaining accurate despite noise inherent in real driving situations. The significance of this work lies in validating the feasibility of deploying intelligent, low-cost IoT sensors for critical vehicle safety applications. By proving that affordable hardware can achieve the precision and real-time performance necessary for rollover risk detection, the study supports the broader adoption of IoT architectures in commercial vehicles. This approach offers a cost-effective alternative to expensive proprietary systems, potentially enhancing road safety by enabling widespread implementation of Roll Stability Control technologies. The findings also provide specific design guidelines for integrating ANNs into embedded systems, contributing to the development of reliable, fault-tolerant vehicular safety networks.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction