Real-time Vehicle Behavior Classification using Single 3-Axis Magnetic Field Sensor and Neural Network

Abstract

This paper presents a real-time method for classifying vehicle behaviors, particularly related to lane violation behaviors by analyzing the output of magnetic sensors. The system assumes the sensor is installed beneath lane markings to detect magnetic field disturbances as vehicles approach. The feature extraction process emphasizes the vertical (Z-axis) magnetic field component and its declination angle, both of which demonstrate robust discriminative characteristics across different vehicle types and positions. A lightweight neural network classification model, based on those features, is trained on these features and deployed on embedded hardware to ensure rapid response and minimal power consumption. The proposed model achieves an overall accuracy of 89.75%, in distinguishing between legal (in-lane) and illegal (lane-crossing) vehicle behaviors. This work introduces a novel integration of simplified signal processing and efficient machine learning suitable for real-time deployment in low-cost Intelligent Transportation Systems (ITS), especially in dense or infrastructure-limited environments.