基于RGB-D数据的物品识别与定位
发布时间：2019-01-17 08:09
【摘要】：近年来,随着智能化进程加快,服务机器人技术得到了飞速发展。物品识别与定位技术作为智能服务机器人必备功能,正逐渐成为研究热点。精准的物品识别和定位能力是机器人执行指定任务的先决条件,在日常生活中有着广泛的应用,如倒水、做饭、拖地等等。传统的物品识别与定位方法多是基于二维RGB图像,根据物品在颜色图像中呈现的诸多特征来进行识别和定位。虽然这些方法取得了不错的效果,但是由于物品的天然特性是三维,无论是形状识别还是6自由度位姿估计都需要充足的空间信息。所以,本文将二维RGB图像数据和三维点云数据相结合,提出了基于RGB-D数据的物品识别与定位方法。文中首先介绍了物品模型数据库的建立过程,通过使用KinectV2传感器分别采集物品的二维和三维数据来构建数据库。然后,在二维RGB图像数据下提取SURF特征,对场景中的物品进行初识别。之后,将初识别的结果映射到三维点云数据下,并使用最小割方法将物品点云从场景中分割出来。最后使用VFH描述子进行物品的精确识别和6自由度位姿估计。此外,为了得到更精确的定位,又提出结合点云配准的6自由度位姿估计方法。实验表明,本方法有效可行。
[Abstract]:In recent years, with the acceleration of intelligent process, service robot technology has been rapid development. As an essential function of intelligent service robot, the technology of object identification and location is becoming a research hotspot. Accurate object recognition and location is a prerequisite for robots to perform assigned tasks, and it has been widely used in daily life, such as pouring water, cooking, mopping and so on. The traditional methods of object recognition and localization are based on two-dimensional RGB images, which are based on many features of objects in color images. Although these methods have achieved good results, due to the natural characteristics of the object is three-dimensional, both shape recognition and 6-DOF pose estimation need sufficient spatial information. Therefore, this paper combines 2D RGB image data with 3D point cloud data, and proposes an object recognition and location method based on RGB-D data. In this paper, the process of building the object model database is introduced, and the database is constructed by using the KinectV2 sensor to collect the 2D and 3D data of the object respectively. Then, SURF features are extracted from two-dimensional RGB images, and the objects in the scene are first identified. After that, the initial recognition results are mapped to 3D point cloud data, and the item point cloud is segmented from the scene using the minimum cut method. Finally, the VFH descriptor is used to identify the object accurately and estimate the position and pose of 6 degrees of freedom. In addition, in order to obtain more accurate location, a 6-DOF position and attitude estimation method of combining point cloud registration is proposed. Experiments show that this method is effective and feasible.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，
本文编号：2409796
[Abstract]:In recent years, with the acceleration of intelligent process, service robot technology has been rapid development. As an essential function of intelligent service robot, the technology of object identification and location is becoming a research hotspot. Accurate object recognition and location is a prerequisite for robots to perform assigned tasks, and it has been widely used in daily life, such as pouring water, cooking, mopping and so on. The traditional methods of object recognition and localization are based on two-dimensional RGB images, which are based on many features of objects in color images. Although these methods have achieved good results, due to the natural characteristics of the object is three-dimensional, both shape recognition and 6-DOF pose estimation need sufficient spatial information. Therefore, this paper combines 2D RGB image data with 3D point cloud data, and proposes an object recognition and location method based on RGB-D data. In this paper, the process of building the object model database is introduced, and the database is constructed by using the KinectV2 sensor to collect the 2D and 3D data of the object respectively. Then, SURF features are extracted from two-dimensional RGB images, and the objects in the scene are first identified. After that, the initial recognition results are mapped to 3D point cloud data, and the item point cloud is segmented from the scene using the minimum cut method. Finally, the VFH descriptor is used to identify the object accurately and estimate the position and pose of 6 degrees of freedom. In addition, in order to obtain more accurate location, a 6-DOF position and attitude estimation method of combining point cloud registration is proposed. Experiments show that this method is effective and feasible.
【学位授予单位】：中国民航大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
，
本文编号：2409796
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2409796.html
