A Path Prediction Model Based on Multiple Time Series Analysis Tools Used to Detect Unintended Lane Departures

Paper i proceeding, 2020

In this paper, a path prediction model is presented and used to detect unintended lane departures caused by erroneous driving behaviours. The prediction model is inspired by the concept of a linear vector autoregressive model that is commonly used for multiple time series analysis. The original concept is extended to allow sparse historic sampling, which is shown to reduce the computational complexity while maintaining the predictive performance. A real world data set is used to derive and validate the proposed model, for which the performance is benchmarked against a kinematic model. The results show that the proposed model can improve the true-positive rate by 18% and reduce the false-positive rate by 34%, with respect to a constant velocity model and for a prediction horizon of 1.75 s.