Convolutional Neural Network for Driving Maneuver Identification Based on Inertial Measurement Unit (IMU) and Global Positioning System (GPS)

Ahmed, Mobyen Uddin; Begum, Shahina · 2020 · Crossref

DOI: 10.3389/frsc.2020.00034

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 challenge of automatically identifying driving maneuvers, specifically navigating roundabouts, to support road safety initiatives and Advanced Driver Assistance Systems (ADAS). While manual identification of maneuvers from Naturalistic Driving Studies (NDS) is labor-intensive, automated systems can help analyze risky behaviors. The research aims to determine the most effective machine learning (ML) algorithm for classifying these maneuvers using data from Inertial Measurement Units (IMU) and Global Positioning Systems (GPS). The authors utilized data collected from 16 volunteers over three months in two countries as part of the EU-funded SimuSafe project. Vehicles were equipped with IMUs (measuring acceleration, pitch, roll, and yaw) and GPS modules. The dataset comprised 76,306 data points, with 847 labeled as roundabouts. Two feature extraction methods were employed: window-based extraction, which used fixed-size windows of eight IMU signals, and maneuver-based extraction, which resampled data and extracted ten features per maneuver, including duration and yaw angle. The study compared six ML algorithms: Convolutional Neural Network (CNN), Hidden Markov Model (HMM), Random Forest (RF), Artificial Neural Network (ANN), K-nearest neighbor (k-NN), and Support Vector Machine (SVM). The CNN was implemented using Keras with specific parameters, including a window size of 9, 1000 epochs, and a class weight of 4:1 to address data imbalance. The results demonstrated that the CNN outperformed the other classifiers in identifying roundabouts, achieving an average F1-score of 0.88 for both roundabout and non-roundabout classes. While the Random Forest algorithm showed a better correlation coefficient (MCC = -0.022), the CNN provided superior overall classification performance. The study highlights that CNNs require less preprocessing and effectively handle the temporal nature of the sensor data. The window-based approach was noted for its stability and lower computation time compared to the maneuver-based method, which suffered from unstable feature values and higher computational costs. The significance of this work lies in establishing CNN as a robust tool for automated driving maneuver identification, particularly for complex maneuvers like roundabouts that are underrepresented in existing literature. By leveraging IMU and GPS data, the proposed system offers a reliable method for analyzing driving behavior, which can contribute to reducing road accidents and improving traffic safety models. The findings support the integration of deep learning techniques in vehicle safety systems and provide a benchmark for future research in automated maneuver detection.

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

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Theoretical Contribution: computational model