SIA OpenIR  > 机器人学研究室
Alternative TitleResearch and verification on road recognition in the field
Thesis Advisor卜春光
Keyword机器视觉 无人驾驶 非结构化道路 Bp神经网络 中值滤波
Call NumberU491/H15/2017
Degree Discipline机械电子工程
Degree Name硕士
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract野外环境中的道路识别技术在灾难救援、军事应用、野外科学考察等领域具有很强的应用前景,现有技术通常只将非结构化道路分成可行区域与非可行区域,而忽视道路本身结构造成的不同道路区域具有不同可通过性。本文主要对非结构化道路的识别问题展开研究,主要完成的工作如下: (1)研究道路模型、车身振动模型,提出将道路不平度作为衡量道路可通过性指标。通过对多自由度简化车身模型的matlab,simulink/car仿真实验,验证该指标的正确性; (2) 研究图像特征提取方法。针对路面不同区域道路不平度不相同的情况,首先对图像进行超像素分割,使得每个超像素内像素颜色、纹理趋于一致,统计每个超像素内所有物理像素的灰度均方根值、面积/周长比、灰度均值、Hue均值、在图像中的高度均值等作为图像中衡量道路不平度的依据; (3) 研究道路不平度特征测量方法。依靠现有设备,分别对使用差分组合导航系统融合激光点云、使用里程计+ICP算法融合激光点云、使用单帧点云测量道路不平度三种方法进行比较,最终确定不进行点云融合只使用单帧点云对道路不平度进行测量; (4) 搭建实验台架,研究激光雷达与相机坐标系之间六自由度标定算法,采集实验数据,使用处理后的数据训练BP神经网络并验证道路提取效果; (5)针对野外环境下蚊虫干扰问题研究经典中值滤波、极值中值滤波、基于排序阈值开关的中值滤波、极大/中值滤波等算法,进行仿真实验,并通过主客观方法评价方法,得到基于排序阈值开关的中值滤波能够取得较好滤波效果的结论。
Other AbstractThe road extraction technology in the field has a strong application prospect in aspect of disaster rescue, military application and scientific investigation. The existing technology usually divides road into feasible and infeasible regions. We always ignore that different road areas have different availability caused by the road structure. In this paper, the problem of unstructured road recognition is studied and the main tasks are as follows: Firstly, we study the road model and car body vibration model. The road roughness is taken as a measure of road performance. Through the simulation multi degree freedom vehicle vibration model in matlab/Simulink and adams/car, we verify the correctness of the index. Secondly, we study the image feature extraction method. Because different road regions have different roughness, the image is divided into superpixels and color and texture of each pixel are consistent. The statistics of the gray RMS value, area / perimeter ratio, gray mean value, Hue mean value and the mean value of height in the image are used as the basis for the measurement of road roughness. Thirdly, we study on the measurement method of road roughness. Based on the existing equipment, three methods are compared:1) the fusion of the laser pointcloud with the GNSS/INS system;2) the fusion of the laser pointcloud with odometer+ICP;3) the single point cloud is used to measure the road roughness. Finally, we use method 3 to measure the road roughness. Fourthly, we build experimental bench and study the calibration algorithm of six degrees freedom between laser and camera coordinate system. The data were collected, and a BP neural network was trained with the processed data. Finally, we verified the effect of road extraction. Fifthly, we study the classical median filtering, extreme median filter, OTSM, maximum median filtering aiming at the problem of mosquito interference in field environment. The simulation experiment is carried out, and subjective and objective evaluation method prove that OTSM can get better filtering effect.
Contribution Rank1
Document Type学位论文
Recommended Citation
GB/T 7714
韩正勇. 野外环境道路识别方法研究与验证[D]. 沈阳. 中国科学院沈阳自动化研究所,2017.
Files in This Item:
File Name/Size DocType Version Access License
野外环境道路识别方法研究与验证.pdf(5519KB)学位论文 开放获取CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[韩正勇]'s Articles
Baidu academic
Similar articles in Baidu academic
[韩正勇]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[韩正勇]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.