Traffic Mode Recognition System Design And Implementation Based On Smartphone

Traffic mode is a kind of the user context. With the development of the intelligent society, recognizing the user’s traffic mode automatically exerts important practical significance. Traffic mode recognition is mainly exploited in the fields of transportation planning and location-based smart services. With the development and popularization of intelligent terminal, the smartphone has been equipped with increasingly powerful abilities of perception, computation, storage and communication. Thus how to recognize traffic mode by smartphone is a hot topic in the current research. The common problem of the current traffic mode recognition research includes single data source, poor adaptive classification algorithm, low recognition accuracy and so on. In this thesis, a traffic mode recognition method based on Random Forest is proposed, and the design and implementation of the traffic mode recognition system is completed based on the method. The achievements are shown as follow.Firstly, sound sensor, gyroscope and public transportation information are introduced for the problem of single data source in this thesis. Sound sensor and gyroscope are rarely used in this research field. The experimental results show that the addition of two kinds of sensors improves recognition accuracy between3.1%and14.7%. Public transportation information is explored for the calculation of public transport-related features. So a user traffic trajectory acquisition method is proposed in this thesis, as well as a calculation method of public transport closeness according to the public transportation information and the user’s trajectory. Public transport-related features characterize the possibility of taking public transportation and improve recognition accuracy by4.2%.Secondly, a traffic mode recognition method based on Random Forest is proposed for the problem of poor adaptive classification algorithm and low recognition accuracy in this thesis. Random Forest is used to build model and select features. Random Forest combination classification algorithm and four kinds of single classification algorithm are compared in this thesis, including Naive Bayes, Bayesian Networks, Decision Trees and Support Vector Machine. The experimental results show that Random Forest algorithm has the best effect. In addition, Random Forest is verified to be superior to the information gain method and reliefF method by experiments in the feature selection research of this thesis.Thirdly, the design and implementation of the traffic mode recognition system is completed based on the method proposed in this thesis. The traffic mode recognition system achieves89.8%accuracy for recognizing eight traffic modes including:static, walking, running, biking, driving, taking a bus, taking a train and taking a subway. The system can not only realize real-time traffic mode recognition, but also provide a friendly user interface for recognition result demonstration combined with map function. Moreover, user historical traffic modes can be stored and displayed by traffic trajectory.