Weigh-in-Motion Method Based on Modular Sensor System and Axle Recognition with Neural Networks

Abstract

Weigh-in-motion systems can measure the number of axles to obtain a vehicle’s type and upper limit of weight, which, combined with the weight measured by the system, can be used for highway toll collection and overload management. This paper proposes a new modular system based on multi-sensor fusion and neural network axle recognition to address issues concerning the high failure rate of axle recognition devices and low weighing accuracy. We use a modular system consisting of multiple weighing platforms, enabling whole-vehicle-load weighing with multiple vehicles traveling through platforms. In addition, we propose a sequential generation model based on a Transformer and Gated Recurrent Unit to calculate the weighing signal generated by the weighing sensors, and then obtain the number of axles and the gross vehicle weight. Finally, the axle recognition algorithm and modular systems are tested in multiple scenarios. The accuracy of the axle recognition is 99.51% and 99.84% in the test set and the toll station, respectively. The weighing error of the modular system in the test field is less than 0.5%, and 99.18% of vehicles had an error of less than 5% at the toll station. The modular system has the advantages of high accuracy, consistent performance, and high traffic efficiency.

Keywords

• axle recognition
• modular system
• transformer
• gated recurrent unit
• weigh-in-motion