基于机器视觉的道路线识别算法研究

发布时间：2018-04-25 04:23

本文选题：机器视觉 + 图像预处理　；参考：《南京理工大学》2017年硕士论文

【摘要】：自上世纪八十年代以来,基于机器视觉的自主导航已成为了智能车辆驾驶技术研究领域的主要方向。车道线识别技术是自主导航的关键技术之一,国内外专家学者在这一技术研究领域做了很多研究,目的是提高识别的鲁棒性和实时性。本文针对较为复杂的道路情况下车道线识别率低、拟合不准确的问题,在保证整个系统实时性的前提下,按照感兴趣区域划分、图像预处理、边缘检测、识别与跟踪的脉络进行如下研究,以提高识别的鲁棒性。文中首先介绍课题研究的背景和意义,并对国内外的研究进行了分析,明确它们存在的问题。其次,对采集的道路图像进行灰度化,再利用垂直灰度均值分布进行初始感兴趣区域分割。预处理阶段分别介绍了图像平滑、增强和边缘检测二值化处理的方法,深入研究了强弱光照下的路面灰度图像处理,并对各种边缘检测算子进行了实验比较,接着设计改进Otsu算法提高了车道线识别度。车道线边缘检测阶段是对车道线的进一步提取,针对噪声对车道线边缘识别的干扰问题,重点提出了分区域识别边缘角度并加以排除的方法去除异常边缘线,并对去噪后的边缘线进行了补偿。车道线识别和跟踪阶段,分析了传统的直线检测和弯道检测方法,并着重对概率Hough变换及RANSAC算法做了改进研究,针对模型灵活性的要求,提出了直线-抛物线型的车道线模型,并设计了模型区域分配的方法以解决曲线道路出现位置不定的情况,再利用最小二乘法求出车道线模型的参数。实验表明,这种方法面对模型不定的结构化道路具有较好的鲁棒性。最后,对得到的初始车道线图像根据其直线模型的斜率与截距,利用Kalman滤波来预测出下一帧的车道线范围,有效的避免过多的噪声干扰。通过对本文算法的仿真实验,证明了本方法具有较好的鲁棒性。
[Abstract]:Since 1980's, autonomous navigation based on machine vision has become the main research field of intelligent vehicle driving technology. Lane recognition technology is one of the key technologies of autonomous navigation. Experts and scholars at home and abroad have done a lot of research in this field in order to improve the robustness and real-time of recognition. In order to solve the problem of low recognition rate of lane line and inaccurate fitting under more complicated road conditions, this paper, on the premise of ensuring the real-time performance of the whole system, divides the area of interest according to the region of interest, image preprocessing, edge detection, etc. The sequence of recognition and tracking is studied as follows to improve the robustness of recognition. This paper first introduces the background and significance of the research, analyzes the domestic and foreign research, and clarifies their problems. Secondly, the road image is grayscale, and then the initial region of interest is segmented by using the vertical gray mean distribution. In the preprocessing stage, the methods of image smoothing, enhancement and edge detection binarization are introduced, and the grayscale image processing of road surface under strong and weak illumination is deeply studied, and various edge detection operators are compared experimentally. Then the improved Otsu algorithm is designed to improve the lane line recognition. The phase of lane edge detection is to further extract lane line. Aiming at the interference of noise to lane line edge recognition, this paper puts forward a method to remove abnormal edge line by recognizing edge angle in different regions and removing abnormal edge line. The edge line after denoising is compensated. In the phase of lane line identification and tracking, the traditional methods of line detection and curve detection are analyzed, and the probabilistic Hough transform and RANSAC algorithm are emphatically studied. According to the requirements of flexibility of the model, a straight-parabola lane line model is proposed. The method of model area allocation is designed to solve the problem that the position of the curve road is uncertain, and the parameters of the driveway model are obtained by using the least square method. The experimental results show that this method is robust to structured roads with uncertain models. Finally, according to the slope and intercept of the linear model, the Kalman filter is used to predict the lane range of the next frame. The simulation results show that the proposed method is robust.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】