基于语义分割的无人驾驶车道线检测算法研究

Research on autonomous lane detection algorithm based on semantic segmentation

自动驾驶车辆在实际行驶过程中,往往对车道线检测的算法效率有着很高的要求,针对这一问题,本文基于2018年提出的经典车道线检测模型Lanenet,提出一种检测时间开销较小的算法,本算法主要通过两个方法提升车道线检测实时性:一是压缩网络结构,减少图片前向处理时间;二是使用时间复杂度小的k-means算法对车道线像素进行聚类,加快聚类效率。在相同实验环境下,本算法在总时间上比Lanenet减少了77.9%,能更好的解决车道线检测实时性能差这一问题。

In the actual driving process of autonomous vehicles, high algorithm efficiency of lane detection is often required. Based on the classic lane detection model Lanenet proposed in 2018, an algorithm with less detection time overhead is proposed. In this algorithm, two methods are used to improve the real-time performance of lane detection: one is to compress the network structure and reduce the time of image forward processing, the other is to use the K-MEANS algorithm with low time complexity to cluster lane pixels, which speeds up the clustering efficiency. Under the same experimental environment, the total time of this algorithm is77.9% less than that of Lanenet. It can better solve the problem of poor real-time lane detection performance.