基于双目视觉的智能车道路环境识别研究

发布时间：2019-01-27 13:28

【摘要】：随着汽车制造业的快速发展以及社会经济水平的普遍提高使得人均汽车持有量逐年升高,汽车已经在全世界范围内越来越普及。汽车行业的快速发展提高了人类的出行半径,使得交通运输更加便利,但是快速发展的汽车工业也给人类和社会带来了很多隐患。比如汽车尾气会对环境造成污染,过多的车辆会使城市交通变得拥挤以及车辆所引发的交通事故等问题。其中交通事故的频发对人类社会安全带来的影响最为严重,如何有效提高车辆行驶的安全性已成为各国科研人员研究的重点内容,因此车辆智能化的概念应运而生。在车辆智能化研究中,采用多传感器(激光雷达、摄像头、探测器)与车辆相搭配的方式是汽车智能化的发展趋势。其中对机器视觉的研究与开发已成为当今研究的重点内容,而双目视觉是机器视觉领域中的重要分支,它是通过模拟生物的视觉来实现机器对环境的感知。因此本文通过应用双目立体视觉的方法来实现对结构化道路上车道线的提取、障碍物的识别、目标距离的测量以及最小安全距离模型的搭建。本文主要研究内容为以下四方面:(1)实验数据采集平台的搭建。本文在实验室电动车上设计并安装一个可移动的水平滑轨,并将双目摄像机安装到滑轨上,其中摄像机可在滑轨上移动并调节不同的基线距离。最终可通过此平台完成对图像的采集。(2)车道线和障碍物检测算法的研究。通过对平台采集到的实验图像进行预处理,然后通过改进的HOUGH变换对车道线进行识别提取。根据检测结果提取出ROI区域,对该区域进行边缘检测以及图像熵值计算找到障碍物的大致区域,再通过SIFT特征算子提取特征点从而找到障碍物的精确位置。对视频序列本文利用Kalman滤波对目标物进行跟踪并结合NMI特征和熵值对预测区域进行验证以实现提高目标跟踪的准确性。(3)基于双目视觉的测距算法研究。在测距过程中采用横向平行式双目视觉系统,根据双目成像原理对采集到的图像进行处理并提取视差值从而完成障碍物距离的计算。并在测距过程中通过大量的实验找到了不同距离下基线长度对测量精度的影响,最终实现降低障碍物测距的误差率。(4)汽车最小安全距离模型的建立。首先对安全预警系统以及汽车制动过程进行分析,然后根据制动原理建立汽车最小安全距离模型,最后在MATLAB中对此模型进行仿真,实验证明此模型达到了预期的效果。
[Abstract]:With the rapid development of the automobile manufacturing industry and the general improvement of the social economic level, the per capita automobile ownership has been increasing year by year, and the automobile has become more and more popular in the world. The rapid development of automobile industry has improved the travel radius of human beings and made transportation more convenient. But the rapid development of automobile industry has also brought a lot of hidden dangers to human beings and society. For example, vehicle exhaust will pollute the environment, too many vehicles will make urban traffic become congested and traffic accidents caused by vehicles and so on. The frequent occurrence of traffic accidents has the most serious impact on human social safety. How to effectively improve the safety of vehicle driving has become the focus of scientific research in various countries, so the concept of intelligent vehicle came into being. In the research of vehicle intelligence, it is the development trend of automobile intelligence to use multi-sensor (lidar, camera, detector) to match with vehicle. Among them, the research and development of machine vision has become the focus of today's research. Binocular vision is an important branch in the field of machine vision, which realizes the perception of machine environment by simulating biological vision. Therefore, the method of binocular stereo vision is used to realize the extraction of lane line, the identification of obstacles, the measurement of target distance and the construction of minimum safe distance model on structured road. The main contents of this paper are as follows: (1) the construction of experimental data acquisition platform. In this paper, a movable horizontal slide track is designed and installed on a laboratory electric vehicle, and a binocular camera is installed on the slide track, where the camera can move on the slide track and adjust the different baseline distance. Finally, the image acquisition can be completed through this platform. (2) the algorithm of lane line and obstacle detection. The experimental images collected by the platform were preprocessed, and then the lane lines were identified and extracted by the improved HOUGH transform. According to the detection results, the ROI region is extracted, and the approximate area of the obstacle is found by edge detection and image entropy calculation. Then the feature points are extracted by the SIFT feature operator to find the exact location of the obstacle. In order to improve the accuracy of target tracking, this paper uses Kalman filter to track the target and verifies the prediction region with NMI features and entropy. (3) the distance measurement algorithm based on binocular vision is studied. The horizontal parallel binocular vision system is used in the ranging process. According to the principle of binocular imaging, the collected image is processed and the visual difference is extracted to complete the calculation of obstacle distance. Through a lot of experiments, the influence of baseline length at different distance on measurement accuracy is found, and the error rate of obstacle ranging is reduced finally. (4) the establishment of vehicle minimum safe distance model. First, the safety warning system and the braking process are analyzed, then the minimum safety distance model is established according to the braking principle. Finally, the model is simulated in MATLAB, and the experimental results show that the model achieves the desired results.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：U463.6;TP391.41

【相似文献】