Lokalizace předmětů pro robotickou manipulaci

Abstract

The new generation of collaborative robots allows the use of small robot arms working with human workers, e.g. the YuMi robot, a dual 7-DOF robot arms designed for precise manipulation of small objects. For the further acceptance of such a robot in the industry, some methods and sensors systems have to be developed to allow them to perform a task such as grasping a specific object. If the robot wants to grasp an object, it has to localize the object relative to itself. This is a task of object recognition in computer vision, the art of localizing predefined objects in image sensor data. This master thesis presents a pipeline for object recognition of a single isolated model in point cloud. The system uses point cloud data generated from a 3D CAD model and describes its characteristics using local feature descriptors. These are then matched with the descriptors of the point cloud data from the scene to find the 6-DoF pose of the model in the robot coordinate frame. This initial pose estimation is then refined by a registration method such as ICP. A robot-camera calibration is performed also. The contributions of this thesis are as follows: The system uses FPFH (Fast Point Feature Histogram) for describing the local region and a hypothesize-and-test paradigm, e.g. RANSAC in the matching process. In contrast to several approaches, those whose rely on Point Pair Features as feature descriptors and a geometry hashing, e.g. voting-scheme as the matching process.The new generation of collaborative robots allows the use of small robot arms working with human workers, e.g. the YuMi robot, a dual 7-DOF robot arms designed for precise manipulation of small objects. For the further acceptance of such a robot in the industry, some methods and sensors systems have to be developed to allow them to perform a task such as grasping a specific object. If the robot wants to grasp an object, it has to localize the object relative to itself. This is a task of object recognition in computer vision, the art of localizing predefined objects in image sensor data. This master thesis presents a pipeline for object recognition of a single isolated model in point cloud. The system uses point cloud data generated from a 3D CAD model and describes its characteristics using local feature descriptors. These are then matched with the descriptors of the point cloud data from the scene to find the 6-DoF pose of the model in the robot coordinate frame. This initial pose estimation is then refined by a registration method such as ICP. A robot-camera calibration is performed also. The contributions of this thesis are as follows: The system uses FPFH (Fast Point Feature Histogram) for describing the local region and a hypothesize-and-test paradigm, e.g. RANSAC in the matching process. In contrast to several approaches, those whose rely on Point Pair Features as feature descriptors and a geometry hashing, e.g. voting-scheme as the matching process.