ABSTRACT
With the highways extending in all directions, the impact of adverse road conditions on traffic safety, mobility and transportation efficiency is becoming even more serious. For the reduction of potential accidents and resource wasting, the road surface condition sensor has become one of the most significant infrastructures for road maintenance departments due to its indispensable role in providing timely road condition information and optimizing maintenance strategy.
Conventional impedance sensors distinguish between ice and water accumulation according to the different impedance response at a specific frequency. However, the differences between the impedance features are relatively unstable when the film of ice or water is too thin to give correct recognition.
In this paper, based on complex impedance spectrum characteristics and machine-learning algorithm, a road surface condition sensor working at the frequency range of 1KHz to 100KHz is proposed and implemented. The direct relationship between the complex impedance and the dielectric property of monitored medium helps to optimize the design of the electrodes sensor. The acquired impedance spectrum curves are processed by the Softmax classifier which inputs 8-dimension features vector and output 4 class labels. The principle and performance of the sensor are investigated in an experimental study and the results indicate the 97.5% recognition correctness of the 1200 validation data.
