Estimation of vehicle trajectories with locally weighted regression

Vehicle trajectory data are important for calibrating driver behavior models (e.g., car following, acceleration, lane changing, and gap acceptance). The data are usually collected through imaging technologies, such as video. Processing these data may require substantial effort, and the resulting trajectories usually contain measurement and processing errors while also missing data points. An approach is presented to the processing of position data to develop vehicle trajectories and consequently speed and acceleration profiles. The approach uses local regression, a method well suited for mapping highly nonlinear functions. The proposed methodology is applied to a set of position data. The results demonstrate the value of the method to development of vehicle trajectories and speed and acceleration profiles. The conducted sensitivity analysis also shows that the method is rather robust regarding measurement errors and missing values.