基于立体视觉的道路场景分割与车辆检测算法研究

发布时间：2018-01-17 03:05

本文关键词：基于立体视觉的道路场景分割与车辆检测算法研究　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着计算机科学和机器人技术的飞速发展,先进驾驶辅助系统(Advanced Driver Assistant System,ADAS)已经成为智能车辆的研究热点,并且广泛应用于军事、民用、科研等相关领域。要保证智能车辆在道路上安全行驶,就要识别道路的可通行区域,也就是避免与道路上的凸障碍物相撞或者陷于凹障碍物中。因此,基于立体视觉的可通行区域与车辆检测算法近年来引起了学术界的关注。本文首先对立体视觉相关领域的研究现状做了简单介绍,然后针对车载立体视觉的道路场景分割与车辆检测,做了以下几个方面的工作:(1)查阅大量文献资料,研究立体视觉的相关理论原理,深刻理解摄像机标定、图像矫正等前期工作,阐述了立体视觉系统的数学模型、成像原理以及U-V视差图的构造,为本文后期工作打下理论基础。(2)搭建移动机器人模拟车载立体视觉系统,编程实现无线手柄对机器人的移动控制,使其能够搭载立体相机在室外移动并采集图像;采用张氏标定法完成摄像机的标定工作,为后续的车辆检测算法提供相机的内外参数;使用SGBM算法对采集到立体图像数据进行立体匹配,并获得稠密准确的视差图;在视差图基础上建立U-V视差。(3)在使用标准霍夫变换对二值化V-视差进行道路建模时,我们经常会遇到的阈值选取困难以及噪声干扰的问题。对此本文引入灰度加权的概念,直接在灰度图像上提取道路特性曲线,从而避免了上述问题并且提高了算法鲁棒性。(4)使用改进动态规划算法检测可通行区域。针对传统动态规划算法检测中会产生的平滑性问题,本文提出改变算法中的平滑项定义,同时考量前后两列的匹配代价,提高了检测准确率;同时对算法提出优化以满足实时性的要求。(5)使用自适应阈值的方法检测障碍物。利用双目相机参数以及障碍物最小高度定义阈值;根据车辆在U-视差图中的投影特征添加阈值化约束条件来确定车辆水平位置,再通过逐列扫描过滤视差值的方法确定垂直位置。实验结果表明本文提出的方法在实际真实世界的各种常见道路环境中都能够区分障碍物区域和可通行区域,并能准确的进行车辆检测,为智能驾驶辅助系统认知外界环境奠定了基础。
[Abstract]:With the rapid development of computer science and robot technology, Advanced Driver Assistant System is an advanced driving aid system. Adas (Intelligent vehicle) has become the research hotspot of intelligent vehicle, and is widely used in military, civil, scientific research and other related fields. To ensure the safe driving of intelligent vehicle on the road, it is necessary to identify the passable area of the road. That is to avoid colliding with a convex obstacle on the road or falling into a concave obstacle. In recent years, the research of passable area and vehicle detection algorithm based on stereo vision has attracted the attention of the academic community. Firstly, this paper briefly introduces the research status in the field of stereo vision. Then aiming at the road scene segmentation and vehicle detection of vehicle stereo vision, we do the following work: 1) consult a lot of literature, study the related theory of stereo vision. The mathematical model of stereo vision system, imaging principle and the construction of U-V parallax map are expounded. For the later work of this paper lay a theoretical foundation. 2) build a mobile robot simulation vehicle stereo vision system, programming to achieve the wireless handle to the robot movement control. Enable it to carry a stereo camera in the outdoor movement and capture images; The camera calibration is completed by using Zhang's calibration method to provide the camera internal and external parameters for the subsequent vehicle detection algorithm. The stereo matching of the collected stereo image data is carried out using SGBM algorithm, and dense and accurate parallax images are obtained. The U-V parallax is built on the basis of the disparity graph.) when the standard Hough transform is used to model the binary V-parallax road. We often encounter the problem of threshold selection and noise interference. In this paper, we introduce the concept of grayscale weighting to extract road characteristic curve directly from gray-scale image. In order to avoid the above problems and improve the robustness of the algorithm. 4) using the improved dynamic programming algorithm to detect the passable area. Aiming at the smoothness problem in the detection of traditional dynamic programming algorithm. In this paper, we propose to change the definition of smoothing terms in the algorithm, and consider the matching cost of the two columns to improve the detection accuracy. At the same time, the algorithm is optimized to meet the real-time requirements. (5) the adaptive threshold is used to detect obstacles. The binocular camera parameters and the minimum height of the obstacle are used to define the threshold. According to the projection feature of the vehicle in the U- parallax graph, a threshold constraint condition is added to determine the horizontal position of the vehicle. The experimental results show that the proposed method can distinguish obstacle areas from passable areas in various common road environments in the real world. And can accurately carry out vehicle detection, for intelligent driving assistance system to understand the external environment laid the foundation.
【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】