非结构化环境下凹障碍的认知技术

发布时间：2018-03-04 09:06

本文选题：移动机器人　切入点：凹形障碍物　出处：《北京信息科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：在环境信息非固定、不可知且结构及尺寸变化不规律、不稳定的非结构化环境下的障碍认知是移动机器人研究领域的关键技术之一,是机器人完成其它高级任务的必要基础。现阶段国内外在非结构化环境中障碍的认知方面已做了很多研究,但是大多针对的是凸起型障碍目标,很少有针对凹形障碍物的研究。本文针对这一问题,以三维激光雷达为主要的视觉传感器,构建了移动机器人平台,提出了一种激光传感多信息特征融合识别算法来完成非结构化环境下有极大威胁、具有一定隐蔽性的沟坑等凹形障碍物的识别,实现非结构化环境下凹障碍的认知。一、构建移动机器人系统平台;对比不同类型的激光雷达,针对本文研究完成了激光雷达选型工作;二、研究点云数据的特点,针对多线激光雷达获得的点云数据提出一种基于曲率分类的滤波方法,以采样点邻域内的平均高斯曲率值来划分点云数据模型的区域类型,再针对不同的区域类型分别采用自适应中值滤波和自适应双边滤波方法,使其既适用于静态目标环境,又适用于动态目标过程;三、针对非结构化环境下的各类障碍进行了检测研究。首先针对水体障碍通过测量激光返回强度值来进行检测,其次基于空间一致性的特点完成了对植被类障碍的检测,最后通过相邻激光束之间的径向间隔的突变来实现对凹障碍的检测。四、针对凹障碍的特点分析移动平台在非结构化环境下的可通过性,针对激光雷达传统直立安装方式在非结构化环境下盲区大、点云数据密度稀疏的缺点,设计一种新的多元激光雷达布局方式,提高在感兴趣区域(ROI)的点云密度,把移动机器人前方盲区的范围控制在1.5米以内;五、提出一种基于贝叶斯准则分配不同权重的激光传感多信息特征融合识别算法来完成在非结构化环境下凹障碍的认知问题研究。针对本文设计的凹障碍认知技术,在所构建的移动机器人平台上进行实验,实验结果表明通过本文的方法可以识别出非结构化环境下2到20米范围内的凹障碍,实现了认知任务。
[Abstract]:Obstacle cognition in unstable unstructured environment is one of the key technologies in the field of mobile robot research, in which environmental information is not fixed, unknown, structure and size change is not regular, and unstable unstructured environment is one of the key technologies in the field of mobile robot research. It is the necessary foundation for robots to accomplish other advanced tasks. At present, many researches have been done on the cognitive aspect of obstacles in unstructured environments, but most of them are aimed at protruding obstacle targets. There are few researches on concave obstacles. Aiming at this problem, a mobile robot platform is constructed based on 3D lidar as the main visual sensor. In this paper, a laser sensing multi-information feature fusion recognition algorithm is proposed to recognize concave obstacles in unstructured environment, such as trenches and pits, which are very dangerous and hidden, so as to realize the recognition of concave obstacles in unstructured environments. Build a mobile robot system platform; compare different types of lidar, for this study completed the type of lidar; second, study the characteristics of point cloud data, A filtering method based on curvature classification is proposed for the point cloud data obtained by multi-line lidar. The mean Gao Si curvature value in the neighborhood of the sample points is used to divide the region type of the point cloud data model. Then adaptive median filter and adaptive bilateral filter are used for different region types to make them suitable for both static target environment and dynamic target process. The detection of all kinds of obstacles in unstructured environment is studied. Firstly, the laser return intensity is measured to detect the water barrier. Secondly, based on the characteristics of spatial consistency, the detection of vegetation obstacle is completed. Finally, the detection of concave obstacle is realized by the abrupt change of radial interval between adjacent laser beams. Fourthly, the transmissivity of mobile platform in unstructured environment is analyzed according to the characteristics of concave obstacle. Aiming at the disadvantages of large blind area and sparse point cloud data density in unstructured environment, a new multi-element laser radar layout method is designed to improve the point cloud density in the region of interest. To control the range of blind areas in front of mobile robots within 1.5 meters; A laser sensor multi-information feature fusion recognition algorithm based on Bayesian criterion is proposed to study the concave obstacle in unstructured environment. The experimental results on the mobile robot platform show that the proposed method can identify concave obstacles within 2 to 20 meters in unstructured environment and achieve cognitive tasks.
【学位授予单位】：北京信息科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【参考文献】