Online intention recognition applied to real simulated driving maneuvers

Abstract

Advanced Driver Assistance Systems (ADAS) are systems developed to assist the human driver and therefore to make driving safer. Research and development of human driving intention recognition (behavior prediction) plays an important role in the development of ADAS, because the estimated driving intention can be used to supervise the drivers intended actions and to avoid dangerous situations. In this contribution, an approach is developed based on fuzzy-Random Forest (fuzzy-RF) for recognizing human driving intentions in real time. Three different driving maneuvers including left/right lane change (LCL/LCR) and lane keeping (LK) are modeled as classes. Driving simulations are generated for highway scenes using a driving simulator. To improve the recognition performance of the proposed approach, membership functions are applied to quantify input signal data into fuzzy sets for RF training. The design parameters of membership functions can be generated automatically by a fuzzy density clustering method. Using experimental data from real human driving, driver intentions are predicted. The results show accuracy values of off-line training phase and on-line test phase are larger than 98 % and 91 % respectively. The effectiveness of driving intention recognition has been successfully proved in this contribution.