Detection of Vehicle Steering Based on Smartphone

Aggressive and abnormal driving may cause serious traffic accidents; therefore, it is crucial and necessary to monitor vehicle steering. For its powerful sensors and processing capacity, the smartphone-based Intelligent Transport System (ITS) can not only capture not only behaviors but also driving styles of vehicle steering. However, vehicle steering behaviors are highly affected by complex conditions such as speed, curvature of the bend, etc. Previous works generally ignore the influence of the speed, and build the models that regard each steering behavior as the same. This paper analyzes the patterns of five common steering behaviors (i.e., left turn, right turn, left lane, right lane and uturn) according to real trace data, and then divide all steering behaviors into 4 speed levels. Different machine learning models are trained on our data set, and the result shows that the decision tree model performs better than other models, and can achieve an average 90% accuracy for the five steering behaviors under four-speed-level.