Real-time Trajectory Prediction Method For Lane-changing Vehicles

This research focuses primary on lane-changing behavior and its safety applications to improve traffic security. As a major part of driving behavior, lane-changing behavior was are reported as a complex and stochastic process comparing to car-following behavior, and also serves as obstacle in collision-algorithm research and corresponding applications. In this paper, a framework of lane-changing trajectory prediction is developed and aim to improve security research on lane-changing behavior. The proposed prediction model is composed by a lane-changing recognition model and a lane-changing trajectory model, and is examined by real vehicular trajectory data. Given sufficient high-solution lane-changing trajectory, factors that affect the lane-changing trajectory in traffic environment was estimated and are exploited in modelling lane-changing trajectory. By examining the impact of driver decision and relative traffic factors on the dynamic process of lane-changing, a formular-based lane-changing trajectory model is finally proposed. In order to recognize the lane-changing behavior in realtime traffic data, a hidden-markov-model-based lane-changing reconginze model is also developed in this pape. Based on the consideration of distinct discrimination between lane-changing and car-following behavior, a series of parameters is selected to form the reconginze model and training process. The reconginze model is capable to detect 90% lane-changing less than 0.5s since the beginning of lane-changing, and is exploit to realtime lane-changing trajectory prediction. As the most important part of prediction model, the procision of lane-changing trajectory model significantly affect the efficiency of prediction model. Factors such as duration of lane-change, velocity, spacing, lane selection, which is affect the trajectory, are glained and estimated in real traffic data, and are applied to describe the lane-changing trajectory. By exaninaming the efficiency of the poposed model, using simulation and real traffic data, the model can generate approximate trajectory comparing to real trajectory: during the lane-changing process, the devation of velocity between real and predict trajection is less than 2m/s in most cases, and at the same time devation of position is less than 3m. Because of the relative long predict time(mean duration of lane-changing is modeled as around 4s), this predict error is relative low. The parameter of the model is easy to glain in real traffic, and result results indicate optimistic performance and will thus benefit safety research on lane-changing.