挖掘机器人多传感器融合的环境感知

发布时间：2018-01-08 10:27

本文关键词：挖掘机器人多传感器融合的环境感知　出处：《西南科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：为了挖掘机器人能够实时的感知野外复杂的环境信息,本文主要从障碍物检测、分割、未知环境地图的建立等几个方面进行了研究。为实现野外环境下对障碍物进行分割,本文提出了一种基于激光雷达信息并结合区域生长的分水岭算法,利用激光雷达信息来做引导分割区域,再将区域生长和分水岭结合起来并利用激光雷达的信息确定区域生长的最优阈值,通过实验验证,该方法有效的克服了分水岭的过分割,并满足实时性的要求。挖掘机器人在作业时,需要对障碍物进行精确的分割,针对Grab Cut耗时的缺陷,本文提出了背景自适应的Grab Cut图像分割算法,即在确定背景像素中选取可能目标像素邻近的一部分像素作为背景像素,减少背景建模的节点数,并利用激光雷达的信息完成交互。实验表明改进的算法有效的提高了运行效率,分割的目标边缘信息更加完整、平滑。由于单目摄像机可视范围无法满足项目需求,本文采用双目图像拼接技术,针对渐入渐出算法融合时出现不清晰等现象,提出了利用Sobel边缘检测处理待拼接的两幅图像的融合区域,然后利用图像边缘差来修正融合区域,实验表明该方法融合后的图像更加清晰。然后详细介绍一种简单的栅格地图的建立方法,利用前帧的障碍物作为路标,在同后帧所建立的局部地图进行匹配更新全局地图,利用激光雷达到后帧图像中的障碍物的距离和角度进行定位;最后验证了基于扩展卡尔曼滤波的SLAM地图构建的可行性。
[Abstract]:In order to mine the robot can perceive the complex field environment information in real time, this paper mainly from obstacle detection, segmentation. In order to realize the segmentation of obstacles in the field environment, a watershed algorithm based on lidar information and region growth is proposed in this paper. The laser radar information is used to guide the segmentation area, then the region growth and watershed are combined and the optimal threshold of the region growth is determined by using the information of the laser radar, which is verified by experiments. This method effectively overcomes the over-segmentation of watershed and meets the requirement of real-time. The mining robot needs to segment the obstacles accurately in the operation, aiming at the Grab Cut time-consuming defects. In this paper, a background adaptive Grab Cut image segmentation algorithm is proposed, in which some pixels adjacent to the possible target pixels are selected as background pixels to reduce the number of nodes in background modeling. The experiment results show that the improved algorithm can improve the efficiency and the edge information of the target is more complete. Smoothing. Because the visual range of monocular camera can not meet the requirements of the project, the binocular image mosaic technology is used in this paper. The fusion region of two images to be stitched using Sobel edge detection is proposed, and then the image edge difference is used to correct the fusion region. Experimental results show that the image fusion is more clear. Then a simple raster map is introduced in detail using the obstacles in the front frame as the signpost. The local map established in the same frame is used to match and update the global map, and the distance and angle from the laser radar to the obstacle in the back frame image are used to locate the map. Finally, the feasibility of constructing SLAM map based on extended Kalman filter is verified.
【学位授予单位】：西南科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】