Robust Lane Detection with Knowledge Distillation

Due to the advancement of artificial intelligence (AI) technology, autonomous driving has recently achieved considerable progress. Lane detection belongs to the perception task which plays an important role of the subsequent task, e.g., planning and decision-making. It involves recognizing the boundaries of driving lanes and should be accurate and real-time. Existing approaches first locate the lane markings in the image and then post-processed it for navigation and safety. This research introduces an end-to-end lane detection approach that incorporates a knowledge distillation-oriented teaching module to forecast the position of the lane. With the help of polynomial regression branch, the teaching module guides the test one capturing the shape features to enhance the lane detection accuracy in challenging road conditions such as complex road layouts. Evaluations conducted on the CurveLanes, CULane, and TuSimple benchmark datasets indicate that our method outperforms the current cutting edge, demonstrating superior accuracy and efficiency.