Research And Implementation Of Traffic Pattern Recognition Based On Intelligent Terminals

In recent years, with the highly popularity of intelligent terminals and the rapid development of mobile Internet, context awareness based on intelligent terminals has become a research hotspot in mobile technology field. And it's an inevitable trend to move the mobile application to the human centered mobile applications. Using the sensing devices embedded in intelligent terminal, we can perceive the real-time location and attitude of current user.In many context-aware research based on intelligent terminal, the detection of the user's traffic mode is one of the core research. It is of great significance for urban transportation planning, promotion of green travel and personalized push service. Current techniques used to detect traffic mode are generally based on GPS, acceleration sensors, GSM or WiFi. But the problems with existing methods are energy consumption,lack of adaptability and stability, or low accuracy because classification features are not comprehensive enough. In addition, the number of types of transportation mode is another challenge to existing methods.Especially the distinction between motor vehicles, due to the uncertainty of road conditions and the difference of driver's driving habits, the distinction between vehicles is difficult to get ideal results.In this paper, we use accelerometer, pressure sensor, gyroscope and magnetic sensor that built-in the intelligent terminals. Through fine-grained analysis of sensor data, the characteristics that can fully reflect the differences between different transportation modes are extracted, Then we use the KNN algorithm to construct walking classifier,use adaboost algorithm to construct static classifier, use the random forest algorithm to construct vehicle classifier, rail vehicle classifier and non-rail vehicle classifier. Finally, a hierarchical classifier consists of five classifiers above is used to detect user's traffic mode. The experimental results show that the accuracy of the proposed method is superior to current GPS-based detection method. Besides, it can avoid relatively large power consumption compared with the GPS detection method, and thus can be better applied to practice. In addition, more effective features can be extracted by observing acceleration and deceleration behavior,stopping behavior and turning behavior during vehicle running.Compared to existing methods based on sensing devices, this method significantly improves the accuracy of motor vehicle category detection,and solves a very challenging problem in traffic mode detection.