| Traffic context recognition is also called traffic mode recognition,which uses the user’s context information to make a recognition and perception of the user’s traffic state.Traffic mode detection is a sub-problem of human behavior recognition.Automatic identification of the traffic mode detection can replace the traditional way of the travel survey,which becomes the most convenient way to obtain a large number of residents travel information data.The data can be used for the city’s transportation planning.Also can be to ease the urban traffic pressure and improve the people’s travel efficiency.This paper explores to use the deep learning methods for modeling the mobile phone sensors to complete the traffic situation identification.First,The paper studies the traditional traffic mode detection method based on mobile sensors,including the type of mobile phone sensor used,data stream processing,and the performance of traditional classification methods.The type of travel modes we will study includes the bus,subway,taxi and high-speed rail.According to the research demand,a benchmark data set of relevant traffic mode recognition is collected and constructed.A total of 255 acquisitions,including 15 collectors,collected 6 different parts,including a total of 7861 minutes of data.In the benchmark dataset,we present two traffic mode detection schemes.The first one is based on the multi-RNN method of travel mode detection.After the sensor is preprocessed,the simple statistical feature is extracted as the input of the RNN network.We use the multilayer or single-layer lstm network to extract timing characteristics for traffic mode recognition.The final recognition accuracy can be 89%;The second method combines The CNN and RNN.This scheme visualizes the sensor data and generate the activity image,then the features of the activity image are automatically extracted by the CNN.The RNN is used to learn the features of the feature images.And finally the recognition accuracy can be 78%. |
PDF Full Text Request