Modeling and analysis for vehicle steering behavior in IPv6 campus environment

Architecture of IPv6-based distributed data collection system and android-based APP to collect vehicle data in a campus environment were proposed, the data communication was based on the WiFi APs, and a vehicle behavior database was built. Through the analysis of different sensor data, the 3-axis of gyroscope was chosen. A MulitWave filter was used for filtering sensor data noise. Finally, a set of eight features was selected as input for training several machine learning-based classifiers. The classifiers were designed for classifying vehicle steering behavior (left/right turn, left/right changing lane and U-turn). Research results show that the decision tree-based J48 classifier works the best and can reach an average accuracy of 96%.