Abstract In this study, we present a perception algorithm that is based purely on vision or camera data. We focus on demonstrating a powerful end-to-end lane detection method using contemporary computer vision techniques for self-driving cars. We first present a minimalistic approach based on edge detection and polynomial regression which is the baseline approach for detecting only the straight lane lines. We then propose an improved lane detection technique based on perspective transformations and histogram analysis. In this latter technique, both straight and curved lane lines can be detected. To demonstrate the superiority of the proposed lane detection approach over the conventional approach, simulation results in different environments are presented.

中文翻译：

---

基于透视变换和直方图分析的自动驾驶汽车车道检测技术

摘要 在这项研究中，我们提出了一种纯粹基于视觉或相机数据的感知算法。我们专注于展示一种强大的端到端车道检测方法，使用现代自动驾驶汽车的计算机视觉技术。我们首先提出了一种基于边缘检测和多项式回归的简约方法，这是仅检测直线车道线的基线方法。然后，我们提出了一种基于透视变换和直方图分析的改进车道检测技术。在后一种技术中，可以检测直线和曲线车道线。为了证明所提出的车道检测方法相对于传统方法的优越性，我们展示了不同环境下的仿真结果。