一种基于双分支车道线实例分割的检测算法

在人工智能的时代,自动驾驶技术越来越成熟,技术中包含的自动车道保持功能占有重要的地位,这对自动驾驶中的后续车道偏离与预警起着关键性的作用。文章利用深度学习技术,针对现有双分支车道线实例分割检测算法存在的准确率受批量影响、准确率不理想等问题,在车道线实例分割中采用自适配归一化函数,并使用传统的SGD优化器对整个模型进行优化解决实验过程中的效率问题。在TuSimple车道数据集进行实验,在性能方面准确率与原始算法相比从96.4%提高到98.6%。

A two-branch lane instance segmentation detection algorithm

In the era of artificial intelligence, autonomous driving technology is becoming more and more mature, and the automatic lane keeping function included in the technology plays an important role, which plays a key role in subsequent lane departure and early warning in autonomous driving. In this paper, deep learning technology is used to solve the problems of existing dual-lane instance segmentation detection algorithms whose accuracy is affected by batches and the accuracy is not ideal. In the lane instance segmentation, an adaptive normalization function is used, and the traditional SGD optimizer optimizes the entire model to solve the efficiency problem during the experiment. Experiments on the TuSimple lane dataset have improved the accuracy rate from 96.4% to 98.6% compared with the original algorithm.