基于立体视觉的室外非结构化环境障碍物检测方法研究
文献类型：学位论文
| 作者 | 彭军舰 |
| 学位类别 | 硕士 |
| 答辩日期 | 2008-06-04 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 沈阳 |
| 导师 | 韩建达 ; 唐延东 |
| 关键词 | 障碍物检测 室外非结构化环境 视差投影图 视差图 |
| 其他题名 | Research on Stereo Vision-based Obstacle Detection in Outdoor Unstructured Environment |
| 学位专业 | 控制理论与控制工程 |
| 中文摘要 | 随着智能机器人系统的发展，机器人的在线感知能力日益受到重视。障碍物检测能力是机器人在线感知能力的一个重要组成部分。因视觉传感器具有独特优势，基于视觉的障碍物检测方法成为目前关注的重点。 室外非结构化环境因结构复杂，机器人缺乏可有效利用的先验知识描述，导致众多障碍物检测系统在该环境中不能有效工作。本文采用全局-局部策略对场景进行由粗到精的分析，弥补室外非结构化环境先验知识不足的难题，提高机器人的在线感知能力。根据该策略，本文在基于视差图的障碍物检测系统框架中，引入视差投影图模块，提出了基于视差投影图的障碍物检测系统框架。该框架在视差投影图模块中全局分析场景视差分布水平，在立体匹配模块中局部分析场景前景目标的几何轮廓信息。依据该框架，针对实际应用中遇到的各种问题，提出了工作于室外非结构化环境的障碍物检测算法。该算法具有如下特点： 1、通过分析视差投影图的地面关联线信息，获得场景的视差分布水平。该信息一方面用来动态更改匹配算法的视差搜索范围，增强算法的实时性和鲁棒性；另一方面用来移除背景地表，简化障碍物分割过程； 2、采用双域滤波抑制噪声，获得清晰的边缘特征，降低立体匹配算法在深度不连续性区域的匹配难度； 3、借助逆向重投影的思想重采样扫描图像，在立体匹配前等效地实现了立体匹配过程中动态变更视差搜索范围的操作； 4、采用基于连通成分的扩散方法替代传统的SAD局部匹配算法，获得高质量的视差图，最终改善障碍物检测的精确性。 在室外非结构化环境中，本文对该算法进行了实验验证。通过设置不同的基线长度，验证了算法在不同的感知距离内的有效性。经实验证明，本算法在一定距离范围内能够有效的检测出障碍物。 |
| 索取号 | TP391.41/P43/2008 |
| 英文摘要 | With the development of the intelligent Robot system, more and more emphasis is focused on robot's ability of on-line environment perception. Obstacle Detection is one fundamental part in Robot perception ability. In terms of vision sensor's unique predominance, vision-based obstacle detection method draws many researchers’ attention. Because outdoor unstructured environment is complicated that robot has no valid description of priori knowledge about this environment, many obstacle detection systems have no perfect performance in this environment. In this thesis, global-local and coarse-to-fine strategies are applied in the perception process, in order to compensate for that lacking prior knowledge about outdoor unstructured environment, and enhance the ability of on-line environment perception. According to the strategies, a module of v-disparity map is introduced into the framework of obstacle detection via disparity map. By this decision, an improved framework is proposed, which is named obstacle detection framework based on v-disparity map. In this framework, the scene is analyzed globally by means of v-disparity map, for obtaining the disparity distribute level, and the geometric contour of scene object is locally analyzed. Adopting this framework, an algorithm for detecting obstacle is designed, which works in outdoor unstructured environment. This algorithm has following features: 1).The Ground Correlation Line is extracted, and the disparity distribute level of current scene is calculated. On the one hand, this information is used to dynamically change the search range of stereo correspondence method. On the other hand, it is used to subtract ground surface, in order to simplify the process of obstacle segment; 2).Bilateral Filter is used to suppress noise, and clearly edge of object in image is obtained. Thus it evidently improves the correct matching rate in depth discontinues region for stereo match; 3).Inspired by Inverse Perception Map, this method dynamically alters search range of stereo correspondence via remapping scan image, which is completed before stereo correspondence; 4).The disparity map is calculated by component Diffusion correspondence method, the accuracy of obstacle detection is also enhanced. In order to validate the performance of our obstacle detection method, we do a series of experiments in different baseline lengths. Experiments show that this method can work on outdoor unstructured environment and have no the phenomenon of omissions. |
| 语种 | 中文 |
| 公开日期 | 2010-11-29 |
| 产权排序 | 1 |
| 页码 | 66 |
| 分类号 | TP391.41 |
| 源URL | [http://210.72.131.170//handle/173321/447]  |
| 专题 | 沈阳自动化研究所_机器人学研究室 |
| 推荐引用方式 GB/T 7714 | 彭军舰. 基于立体视觉的室外非结构化环境障碍物检测方法研究[D]. 沈阳. 中国科学院沈阳自动化研究所. 2008. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。