Research On The Behavior Sensing And Classification Of Mobile Vehicle Based On Smart Devices

In recent years,with the development of economy and technology,urban traffic problems have become increasingly prominent.The construction of ITS(intelligent trans-portation system)which can be convenient for people's lives and solve the increasingly serious traffic problems has become an urgent demand.One of the important problem in the construction of ITS is the mobile vehicle's behaviour perception.At present,smart phones,smart watches and other smart device equipment has become necessities in our daily life.They are equipped with a variety of sensors with a strong perception capability of the external environment or dynamic information.This paper focused on the problem that using the inertial sensors built in mobile devices to sense the vehicle's dynamic be-haviors.Meanwhile,using the unsupervised learning method to identify and classify the different behaviors.The whole paper 's contributions are as below:(1)For the data that collected by inertial sensor(acceleration sensor,gyroscope),we made quantitative analysis and determined a proper sample frequency.Meanwhile,we made spectrum analysis for the data to design appropriate filter in order to filter the noise,which can make the sensor data more accurately.(2)Using multi-sensor data fusion method to determine the attitude of smart devices,and then determine relation between the smart device's coordinate and vehicle's.In the multi-sensor data fusion method,we compare the complementary filter and Kalman fil-ter.From the result,we choose to use Kalman filter as a data fusion method to get the attitude of the device.And then determine the correspondence between the smart device's coordinate and the vehicle's according to the attitude information.(3)The physical model of the common behavior of the vehicle was analysed.Ac-cording to the physical model analysis,The statistical characteristics of the sensor data from the inertial sensor can be extracted to reflect the vehicle's dynamic behavior.These multi-dimensional data features can be further used as a principle for the vechicle's be-havior's identification and evaluation.In this way,we can avoid past work's limitation that using thresholds to judge different behavior.(4)Based on the statistical characteristics of the sensor data collected in the re-al urban environment,we identify and classify the common behaviors(lane change,left turn,right turn,U turn)of the vehicle.A new method based on multi-feature comprehen-sive score is designed to determine the behavior of lane change.For the other dynamic behaviors5 classification that are common in everyday life,such as left turn,right turn,U-turn behavior was based on the unsupervised learning classification method.We use the K-means method to classify these behaviors by making full use of the statistical char-acteristics of inertial sensor data.Then we evaluate the result of different kinds of mobile platform's behavior identification and classification.The result shows that our approach is effective and accurate.