Lane Environment Detection with Only On-Board Camera

Lane environment detection is a vital prerequisite for realizing autonomous driving in complex scenarios. However, existing methods face significant challenges, primarily containing the absence of lane-level positioning information, the imbalance between accuracy and processing speed, and the prohibitively high cost of equipment. To address these challenges, this paper introduces a hybrid method that combines deep learning and traditional image processing techniques. The proposed approach uses the YOLO (You Only Look Once) for environmental vehicle detection, augmented by a post-processing module that provides detailed lane-level positioning information. Notably, the post-processing stage integrates deep learning and traditional computer vision methods, ensuring a fast detection speed while maintaining a desirable detection accuracy. An additional advantage of the proposed method is its reliance solely on an on-board camera, effectively lowering the equipment cost. Furthermore, the method provides more comprehensive detection by incorporating geometric method to estimate the distance between the ego-vehicle and surrounding vehicles. The proposed approach addresses the aforementioned challenges and provides a comprehensive solution for effective and low-cost lane environment detection. To be specific, our proposed method runs at 58 fps and achieves accuracy at 95.22%.