Vehicle in Motion Weighing Based on Vibration Data Collected from Sensor Network

Abstract

This paper introduces a method for estimating weight of moving vehicle based on vibration measurements registered by sensor network. The sensor network consists of sensor nodes installed on roadside and curbs. The sensor nodes calculate statistical features of vibration measurements and verify them using a proposed quality function. Then statistical features are used to create a vibration model, which allows the sensor nodes to estimate vehicle weight. Vibration models were created in this study using linear regression and fully connected neural network. A final assessment of vehicle weight is performed at the network sink, by taking into consideration the estimations and data quality indicators provided by particular sensor nodes. Several machine learning methods were compared with the proposed approach. The proposed method has low computational requirements, thus it can be adapted to sensor networks with computational and memory constrains. Advantages of the introduced method were demonstrated in real-world experiments, using various modelling approaches and installation types of sensor nodes. The experimental results confirm that the introduced method can be adapted for the node, which can weight vehicles independently or in cooperation with other nodes (as an ensemble).