Prediction Performance of Lane Changing Behaviors: A Study of Combining Environmental and Eye-Tracking Data in a Driving Simulator

Abstract

Advanced Driver Assistance Systems (ADAS) are systems developed to assist the human driver and therefore to make driving safer and better. Understanding and predicting human driving behavior play an important role in the development of assistance systems. In this contribution, the development of a driver assistance system is based on the prediction of driving behaviors. The driving patterns of three different behaviors are modeled including left/right lane change and lane keeping. A driving simulator is used to simulate a highway scene. The implementation of a prediction system based on different machine learning approaches such as Hidden Markov Model (HMM), Support Vector Machine (SVM), Convolutional neural networks (CNNs), and Random Forest (RF) is accomplished. In addition, eye-tracking information is integrated. The task is to predict behaviors based on the measurement. As test, a 10-fold cross-validation is used based on data sets from driving simulator and applied to compare the performance of different algorithms. In combination with related results in terms of accuracy (ACC), detection rate (DR), and false alarm rate (FAR), the performance and effectiveness of the developed prediction systems are evaluated. The results show that the performance of RF algorithm is the best of all four algorithms compared. Combining environmental and eye-tracking data the RF algorithm achieved the best results. All ACC values are larger than 99 %. Afterwards, two RF-based prediction models with and without eye-tracking data are developed for online test. Finally, some application samples are suggested for driver assistance. The results calculated by the proposed model are shown on a user interface to help the drivers to see when it is suitable to turn left, to turn right, or to keep the direction.