基于机器视觉的汽车行驶道路状况感知技术的研究

发布时间：2018-06-14 19:41

本文选题：机器视觉 + 道路状况感知　；参考：《华南理工大学》2014年硕士论文

【摘要】：近年来世界范围内尤其在中国汽车保有量持续增加，但是越来越多的道路汽车驾驶安全隐患也在增长，故对提高安全驾驶的辅助驾驶系统需求越来越强烈。本文中基于机器视觉的汽车行驶道路状况感知系统正式辅助驾驶系统的一部分。通过本系统可以对单目摄像机采集的图像进行分析处理，并得到道路车道线的位置信息与车辆的位置信息，通过这些信息对辅助驾驶系统的反馈，很大程度上可以提高道路汽车驾驶的安全。首先，本文引入了两项道路智能感知系统的关键技术：车道线检测和车辆检测。并对机器视觉的概念与两项关键技术前的图像预处理进行阐述。其次，研究了用于车道线检测的随机霍夫变换算法，，考虑到该算法的准确性与实时性问题，采用了区域约束与角度约束的方法提高了车道线检测的准确性与实时性。然后，对车辆检测方法进行深入研究，根据车辆检测准确性与实时性的需求，并整理出一套较好解决高准确性与实时性需求的方法。将检测过程分为了假设产生模块与假设验证模块。针对假设产生模块对实时性的优化与假设验证模块对准确性的优化，分别详细设计了两个模块的各自框架与实现方法，包括基于AdaBoost和Haar-like特征的假设产生框架与基于SVM和HOG特征的假设验证框架，以上两个框架分别结合了级联AdaBoost快速筛选与Haar-like特征快速计算的速度优点以及SVM对高维特征的分类与HOG特征丰富的梯度信息的准确性优点。经系统实现与测试验证，本文研究的汽车行驶道路状况智能感知系统，在结构化道路情境下，达到了较高的检测率与较优的实时性。
[Abstract]:In recent years, the number of automobile ownership has been increasing in the world, especially in China, but more and more hidden dangers of driving safety are also increasing, so the demand of auxiliary driving system for improving safe driving is becoming more and more intense. This paper is a part of the vehicle driving condition perception system based on machine vision. Through this system, the image collected by the monocular camera can be analyzed and processed, and the position information of the road lane and the vehicle can be obtained, and the feedback to the auxiliary driving system can be obtained through these information. To a large extent, it can improve the safety of driving on the road. Firstly, this paper introduces two key technologies of road intelligent sensing system: lane detection and vehicle detection. The concept of machine vision and the image preprocessing before two key technologies are expounded. Secondly, the random Hough transform algorithm for lane detection is studied. Considering the accuracy and real-time of the algorithm, the method of region constraint and angle constraint is adopted to improve the accuracy and real-time performance of lane detection. Then, the method of vehicle detection is studied deeply, according to the demand of accuracy and real-time of vehicle detection, and a set of methods to solve the requirement of high accuracy and real-time is put forward. The detection process is divided into hypothesis generation module and hypothesis verification module. Aiming at the optimization of the real-time performance of the hypothesis generation module and the optimization of the accuracy of the hypothesis verification module, the respective frameworks and implementation methods of the two modules are designed in detail. The hypothesis generation framework based on AdaBoost and Haar-like features and the hypothesis verification framework based on SVM and HOG features are included. These two frameworks combine the advantages of cascaded AdaBoost fast filtering and Haar-like feature fast computation and SVM classification of high-dimensional features and the accuracy of hog feature rich gradient information. Through the system realization and test, the intelligent sensing system of vehicle driving road condition is studied in this paper. Under the situation of structured road, the intelligent sensing system achieves higher detection rate and better real-time performance.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;U495

【参考文献】