Real-Time Prediction of Multi-Class Lane-Changing Intentions based on Highway Vehicle Trajectories

For fully automated driving in the real-world, safety, comfort, and reliability are the key essentials during operation. However, in the automotive industry, it is critical to identify the lane change intention of vehicles from the adjacent lane for safe driving. Thus, the intelligent assistance system should be smart enough to track such lane change desires and assist the driver to avoid any accidents that is caused by human errors on the road. Here, in this paper we propose to predict the driving behaviors of lane-changing vehicles on the highways by studying their chronology. Machine learning performs well in retaining the human-vehicle behavioral patterns and therefore, we involved random forest to do this prediction. We proposed new feature space for lane change prediction, where a really small window size of 3-seconds can lead to high-performance. The extensive experiments are carried out on real-traffic dataset from Next Generation SIMulation (NGSIM) where we set up a new state-of-the-art accuracy of 98.6%.