基于实例分割方法的复杂场景下车道线检测

The lane line detection in complex scene based on instance segmentation

面对当前在复杂场景下车道线检测存在精度不高、鲁棒性较差等问题,提出一种基于实例分割方法的复杂场景下车道线检测算法。该算法基于改进的VGG-16卷积神经网络结构,首先对道路图像进行二值化语义分割,得到离散的车道线像素点;再采用Meanshift聚类方法确定属于同一条车道线的像素点,形成相应的车道线实例;最后结合可变透视变换矩阵变换道路图像,采用多项式拟合生成车道线参数方程。该算法能够在车道线残缺、被阴影遮挡或数量发生变换等复杂场景下,实现对车道线的准确检测,具有较高的识别准确率。实验结果表明,该算法在不同场景下弯道和直道检测的平均准确率为96.6%,能适合多种路况车道线检测,鲁棒性较好。

Aiming at the current low precision and poor robustness of lane detection in complex scenes, it proposes an algorithm based on segmentation method for lane detection in complex scenes. The algorithm is based on the improved VGG-16 convolutional neural network structure. Firstly, the road image is binarized and semantically segmented to obtain discrete lane line pixels. Then the Mean shift clustering method is used to determine the pixels belonging to the same lane line. The corresponding lane line instances are formed;finally, the road image is transformed by the variable perspective transformation matrix, and the lane line parameter equation is generated by polynomial fitting. The algorithm can accurately detect lane lines in complex scenes such as lane line defects, shadow occlusion or quantity transformation, and has high recognition accuracy. The experimental results show that the proposed algorithm has an average accuracy of 96.6% for corner and straight track detection in different scenarios, which can be used for lane detection of multiple road conditions with good robustness.