کنفرانس مهندسی برق ایران

صفحه اصلی / سی و چهارمین کنفرانس بین المللی مهندسی برق

Multi-Sensor Data Fusion for Real-Time Road Surface Type and Condition Classification Using Wheel-Mounted Accelerometers and Weather Sensors

نویسندگان :

Nima AhmadPour¹

1- minho university

کلمات کلیدی :

Real-time road surface recognition،Multi-sensor data fusion،Wheel-mounted accelerometers،Discrete wavelet transform،Embedded machine learning

چکیده :

Accurate real-time identification of road surface type and condition is essential for intelligent vehicle control and advanced driver assistance systems. This paper presents a novel multi-sensor data fusion framework for joint classification of road surface type and surface condition under real-world driving scenarios. The proposed system integrates vibration signals from four wheel-mounted low-cost accelerometers manufactured by Bosch with environmental measurements, including humidity and road water film thickness, obtained from a front-mounted weather sensor module (MD30) by Vaisala. Accelerometer signals are segmented and processed using discrete wavelet transform, and statistical features are extracted from multi-scale approximation and detail coefficients. These vibration-based features are fused with environmental features of different physical nature, forming a unified feature representation. A multi-class support vector machine is trained using combined labels that simultaneously encode road surface type and condition (e.g., asphalt–dry, dirt–wet), with hyperparameters optimized via cross-validation. The complete machine learning pipeline is implemented and deployed on an in-vehicle embedded platform for real-time inference, and the inferred road information is designed to be transmitted to a cloud server for collective road awareness. Experimental results obtained from extensive data collected across multiple locations, speeds, driving modes, and drivers demonstrate a classification accuracy of 98.5\%, confirming the effectiveness and reliability of the proposed heterogeneous feature fusion approach. The results highlight the strong potential of the proposed framework for scalable and cost-effective intelligent vehicle applications.