Real-time lane prediction algorithm using the sequence-to-sequence model

Abstract

In this research, we propose a real-time lane prediction algorithm using the sequence-to-sequence model. Deep learning based on camera sensors has become an indispensable key technology for situation recognition while playing the same role as the eyes of autonomous vehicles. However, various situations on the actual road make many restrictions on the cameras to recognize the situation. In particular, various situations such as a section where the lane is cut off such as a crosswalk or a crossroad, a situation where the lane is cut off due to the painted state of the lane on the road, and a situation where the lane is blocked by other vehicles or obstacles on the road increase the risk and instability in autonomous driving. In this research, we propose the real-time lane prediction algorithm based on the detected lanes after acquiring a dataset by driving an actual autonomous vehicle.