走行車のタイヤ音を利用した路面状況の検出に関する研究

摘要

Information on road surface states is important and helpful for road users such asautomobile drivers, particularly in the snowy season. In practice, the surface statesdepend greatly on the weather, road users, location, and other relevant factors. Thisdissertation is concerned with the reliable detection of the surface states using tire noisefrom road vehicles. The tire/road noise emitted from moving vehicles variesmomentarily depending on the road surface properties. Then, it may be possible topassively and easily detect the state of the road surface: i.e., dry, wet, snowy, or otherstate. To detect tire noise, the author used a commercially available microphone as anacoustic sensor, which enabled us to easily reduce the cost and size in realizing apractical system for detecting road surface states. Several features in the frequency andthe time domain of noise signals are proposed to successfully classify the road statesinto several categories and to improve the classification accuracy by combining theindicator of the feature obtained with the standard deviation of the cumulativedistribution curves. Furthermore, this dissertation proposes a new processing methodfor automatically detecting the states of road surface. The method is based on artificialneural networks. The proposed classification is carried out in multiple neural networksusing learning vector quantization. The outcomes of the networks are then integrated bythe voting decision-making scheme. From the experimental results obtained in snowyseason demonstrated that an accuracy of approximately 90% can be attained forpredicting road surface states using only tire noise data.