SIA OpenIR  > 机器人学研究室
机器人视觉导航中的实时在线识别算法研究
Alternative TitleReal-time Object Recognition for Robot On-line Navigation Based on Vision
丛杨1,2
Department机器人学研究室
Thesis Advisor唐延东
ClassificationTP242.62
Keyword计算机视觉 模式识别 目标识别 自主导航 机器人
Call NumberTP242.62/C87/2009
Pages100页
Degree Discipline模式识别与智能系统
Degree Name博士
2009-05-25
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract随着移动机器人应用范围的日益扩展,在动态、非结构化环境下提高其自主导航能力已经成为移动机器人研究领域迫切需要解决的问题。在机器人自主导航关键技术中,识别技术是最难解决、也是最急需解决的问题。视觉作为导航中的重要传感器,与其他传感器相比具有信息量大、重量轻便、功耗低等诸多优势,因此基于视觉的识别技术也被公认为最具潜力的研究方向。本文以国防基础研究项目和中科院开放实验室基金项目为依托,以沈阳自动化所自主研发的“轮腿复合结构机器人”和“无人机”为实验平台,针对地面自主机器人和无人机自主导航中迫切需要解决的应用问题,有针对性的展开研究,旨在提高移动机器人在动态、非结构化环境下的适应能力。本论文的主要内容如下: 首先,为了提高复杂环境下地面移动机器人的自主能力,本文提出了一种基于立体视觉的面向室外非结构化环境障碍物检测算法。文中首先给出了一种可以从V视差图(V-disparity image)中有效估计地面主视差(Main Ground Disparity, MGD)的方法。随后,我们利用由粗到精逐步判断的方式,来识别疑似障碍和最终障碍并对障碍进行定位。最后,该方法已在地面自主移动平台得到实际应用。通过在各种场景下的实验,验证了该方法的准确性和快速性。其次,以无人机天际线识别为背景,提出了一种准确、实时的天际线识别算法,并由此估计姿态角。通过对天际线建立能量泛函模型,利用变分原理推出相应偏微分方程。在实际应用中出于对实时性的考虑,引入分段直线约束对该模型进行简化,然后利用由粗到精的思想识别天际线。具体做法是:首先,对图像预处理并垂直剖分,然后利用简化的水平直线模型对天际线进行粗识别,通过拟合获得天际线粗识别结果,最后在基于梯度和区域混合开曲线模型约束下精确识别天际线,并由此估计无人机滚动和俯仰姿态角。第三,通过对红外机场跑道的目标特性进行分析,文中设计了一种新的基于1D Haar 小波的并行的红外图像分割算法的;然后,有针对性的对分割区域提取特征;最后,两种常用的识别方法,支持向量机(SVM)和投票法(Voting)被用于对疑似目标区域进行分类和识别。通过对实际视频和红外仿真图片的测试,验证了本文算法的快速性、可靠性和实时性,该算法每帧平均处理时间为30ms。最后,针对无人机空中巡逻中对人群进行自动监控所遇到的问题,通过将此类问题简化为固定视角下人流密度监测问题,提出了一种全新的基于速度场估计的越线人流计数和区域内人流密度估计算法。 首先,该算法把越线的人流当成运动的流场,给出了一种有效估计1D速度场的运动估计模型;然后,通过对动态人流进行速度估计和积分,将越线人流的拼接成动态区域;最后,对各个动态区域提取面积和边缘信息,利用回归分析实现对人流密度估计。该方法与以往基于场景学习的方法不同,本文是一种基于角度的学习,因此便于实际应用。
Other AbstractAutonomous navigation is one of the most critical issues for mobile robot in dynamic and unstructured environment, which are the necessary performance demanded by real application. Because the surrounding environment is high degree of uncertainty and the goal is not completely predictable, the recognition algorithm becomes one of the most difficult technologies for robot autonomous navigation, and it is also the most urgent problem. Vision,as an important navigation sensors, has many advantages compared to other sensors, such as the most amount of information, light weight, low power consumption, etc. Therefore, the Vision-based navigation is the most potential research topic in robot navigation. In this dissertation, all of the recognition algorithms are intended to the Unmanned Ground Vehicle and Unmanned Aerial Vehicle platofrm developed by State Key Laboratory of Robotics, Shenyang Institute of Automation, P.R.China. Firstly,a fast obstacle detection (OD) system based on stereo vision for Unmanned Ground Vehicle (UGV) navigation in unstructured environment is presented. In order to make the UGV adaptable to more complex terrains, a new estimation method of the Main Ground Disparity (MGD) from the V-disparity images is proposed. Then by comparing the disparity of the MGD with local 3D reconstruction, a Coarse-to-Fine method to find and localize obstacles is introduced in the paper. The obstacle detection system is tested practically on our UGV platform in some outdoor unstructured environments. The experimental results validate our system. Secondly,an accurate, real-time skyline detection algorithm on video frequency for Unmanned Aerial Vehicle (UAV) is presented. By curve evolution, a skyline energy functional model is established firstly. In practical application, the linear constraint is adopted to simplify the model to reduce computing time, and then coarse-to-fine strategy is used to extract the skyline. The rough local skyline is detected by a group of short horizontal lines, and then the rough skyline is fitted by the rough local skylines. At last, the gradient based model is used to detect the accurate skyline, and the rolling angle of the UAV can be calculated by the final detected skyline. The results of the experiment show that our skyline detection algorithm has robust, accurate and real-time performance. Thirdly, a real-time Automatic Target Recognition (ATR) algorithm for infrared runway targets under complex backgrounds is proposed. By analyzing the features of the infrared runway targets, a novel parallel infrared image segmentation algorithm based on 1D Haar wavelets is proposed. The feature descriptors, which can properly denote the shape and regional features of the infrared runway targets, are designed and these features are extracted. By comparison with the Voting algorithm, the Support Vector Machine (SVM) is utilized to train the regional features and to recognize the targets. The experiments show that our recognition algorithm, especially for the runway targets, is accurate, robust and real-time. The average time consumption of our algorithm is up to 30ms/frame. In the end, a novel algorithm based on flow velocity field estimation to count the number of pedestrians across a detection line or inside a specified region is proposed. Pedestrians across the line are regarded as fluid flow, and a novel model to estimate the flow velocity field is proposed. By integrating over time, the dynamic mosaics are constructed to count the number of pixels and edges passed through the line. The number of pedestrians is estimated by quadratic regression, with the number of weighted pixels and edges as input. The regressors are learned off line from several camera tilt angles with the calibration information. The tilt-angle-specific learning is used to ensure direct deployment and avoid overfitting while the commonly used scene-specific learning scheme needs on-site annotation and sometimes result in overfitting. Experiments on a variety of videos verified that the proposed method can give accurate estimation under different camera setup in real-time.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/504
Collection机器人学研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
丛杨. 机器人视觉导航中的实时在线识别算法研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2009.
Files in This Item:
File Name/Size DocType Version Access License
10001_20061801472803(13144KB) 开放获取--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[丛杨]'s Articles
Baidu academic
Similar articles in Baidu academic
[丛杨]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[丛杨]'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.