Research And Implementation Of Traffic Status Recognition Based On Mobile Phone Sensor

With the rapid development of cities,traffic congestion has become one of the key issues facing major cities.Accurately identifying and predicting real-time traffic conditions is of great significance in alleviating traffic congestion,building urban intelligent transportation systems,and promoting modern social development.At present,traffic state recognition methods and research mainly rely on traditional equipment such as on-board GPS and fixed monitors.Such equipment has the disadvantages of high installation and maintenance costs,small coverage area,and large energy consumption,which makes traffic state recognition have many problems in practical application.With the popularization of smart phones and the continuous development of sensor technology,the use of rich types of mobile phone sensors for traffic state recognition has the advantages of low cost,low power consumption,and high stability.It is a new idea to effectively solve the shortage of traditional detection equipment.Based on this,thesis studies traffic state recognition based on mobile phone dynamic sensors.The main research contents include the following four aspects:(1)Construction of traffic state data set based on mobile phone sensorsIn order to obtain the sensor data efficiently and conveniently,a data collection tool is designed to collect the sensor data generated by the vehicle during driving and mark it.Secondly,through the data pre-processing and analysis of the original data and its time-domain features,the feature data used to identify the traffic state model was selected to build the data set required for subsequent research.(2)Traffic state recognition based on SVMThe traffic state recognition problem is a multi-class non-linear classification problem,considering the excellent performance of SVM for such problems,a traffic state recognition model based on SVM is constructed.Firstly,different kernel functions and characteristic combinations are compared.The experimental results show that when using the mean and variance time-domain features of the acceleration sensor and gyro sensor data as the feature data,the SVM model of the RBF kernel function has the best recognition effect.In order to further improve the model recognition effect,an improved ABC algorithm optimized parameter SVM traffic state recognition model is proposed.The experimental results show that the accuracy of the SVM model optimized by the improved ABC algorithm reaches 91.66%,which is 3.25% higher than the SVM model of traditional gridsearch method.(3)Traffic state recognition based on LSTMBased on LSTM network,it is good at processing time series data,strong nonlinear processing and generalization ability.In order to improve the effect of traffic state recognition,LSTM-based traffic state recognition is further studied.After performing pre-processing operations such as data format conversion on traffic status data,first use Adam's gradient descent method to compare experiments to select Adam to update model parameters,and then based on the real traffic status data set,experiment and verify that the LSTM model in thesis has good traffic status recognition effects The accuracy rate is 92.67%,which is 1.82% higher than the SVM model optimized by the improved ABC algorithm.(4)Design and implementation of traffic status recognition systemBased on the LSTM traffic state recognition model,thesis finally implements a lightweight Android software with functions with functions of data collection and marking,traffic status recognition,traffic status monitoring and so on,as a preliminary exploration of mobile phone sensor traffic state recognition applications.