A Predictive Model of a Driver's Target Trajectory Based on Estimated Driving Behaviors

With the development of automated driving, inferring a driver's behavior can be a key element for designing an Advanced Driver Assistance System (ADAS). Current research is focused on describing and predicting a driver's behaviors as labels, e.g., lane shifting, lane keeping, etc., during driving. In our work, we consider that predicting a driver's behavior can be described as predicting a trajectory the driver may follow in the near future. The target trajectory can be calculated through certain polynomial functions. Via the data set collected by a Driving Simulator experiment covering nine volunteers, we proposed a model based on a deep learning network which is capable of predicting the corresponding coefficients of polynomial functions and then generating the trajectories in the next few seconds. The results also discussed and analyzed some possible factors affecting the prediction error. In conclusion, the model proved to be effective in predicting the target trajectory of a driver.