基于IPM和边缘图像过滤的多干扰车道线检测

在基于视觉的自动驾驶环境感知中，路面阴影、雨水、污渍和反光会对车道线识别和车辆导航造成干扰，针对此问题提出了一种基于逆投影映射（IPM）和边缘图像过滤的改进车道线识别方法。通过逆投影方法可以得到原始道路图像的鸟瞰图像，很大程度上增强了车道线的视觉特性并减少了干扰。同时提出迭代聚类分割方法对IPM图像中的灰度值进行分析，并保留与车道线颜色和形态特征最为接近的灰度点作为车道线边缘。随后提出一种搜索统计边缘图像中连续边缘区域的方法，通过分析边缘点并保留最长区域实现过滤道路干扰因素的目的。最后将该算法与其他常用车道线检测算法进行对比。研究结果表明：该方法可以更好地过滤路面各种干扰因素，有效增强干扰环境下识别模糊车道线、实车道线、虚车道线、弯车道线的能力，大幅提高了自动驾驶环境中的车道保持能力，并且由于该方法相比其他方法能够更加有效地去除路面干扰区域，因此识别车道线的速度得到大幅提高，可以满足自动驾驶对于实时性的要求。

Multi-interference Lane Recognition Based on IPM and Edge Image Filtering

In visual-based environment perception for autopilots, shadows, stains, water, and reflected light could interfere with lane recognition and navigation. In this paper, an improved lane recognition algorithm based on inverse perspective mapping (IPM) and edge image filtering was proposed to solve this issue. An aerial image of the original road scene could be obtained through IPM, which would significantly enhance the visual characteristics of the lane and reduce interference. An iterative clustering segmentation method was proposed to analyze the gray values of the IPM gray image, and the gray points closest to the color and morphological features of the lane were retained as the lane edge in the IPM image. Subsequently, a method that could search and determine the statistic of continuous edge regions was developed to segment the edge image. Filtering the interference factors was achieved by analyzing the edge points and retaining the longest regions. In comparison with other commonly used lane detection algorithms, the result indicates that our method can more effectively filter all kinds of interference factors on the road and enhance the ability to detect fuzzy, real, virtual, and curved lanes under an environment with interference. This significantly improves the ability to keep to a lane under an autopilot environment. Because of this, lane recognition speed is greatly improved, which can meet the real-time requirements autopilots.