3D Scene Reconstruction Based On Cloud Robotics

3D scene reconstruction is an important research field of robotics, which played an important role in many robotic applications. The goal of 3D scene reconstruction is to explore the unknown environment, and to establish the 3D model of the environment with different sensors. But the process of 3D scene reconstruction depends on the CPU, memory, hard disk and other hardware configuration of robot heavily. The robot with low configuration can’t meet the condition of 3D scene reconstruction process which needs massive data storage and intensive computing. The arise of Cloud Robotics provides a new approach to this problem. This paper based on the cloud platform established in the laboratory, researches the method of 3D scene reconstruction technology with Kinect and proposes a 3D scene reconstruction architecture based on the cloud platform. The main works of this paper are divided into the following sections:Firstly,3D scene reconstruction methods are studied in this paper. Through the comparison of SIFT, SURF and ORB matching algorithm, this paper chooses ORB algorithm to match RGB images, and uses RANSAC algorithm to eliminate false matching. This paper employs the ORB algorithm and the RANSAC algorithm to get the feature matching pairs of 2D images captured by Kinect, then maps these feature pairs to 3D point cloud combined with the depth image. This paper uses the SVD algorithm to calculate the rotation matrix and translation matrix between the point clouds, then uses a nonlinear least squares optimization algorithm LM algorithm for loop closure detection of multi-field spots cloud data to decrease accumulative error due to clouds stitching. By the methods, the multi-scene point cloud data can be effectively splitted.Secondly, we design and implement the architecture of single robot 3D scene reconstruction based on the cloud platform. Combined with the idea of Cloud Robotics, this paper designs the architecture of 3D scene reconstruction based on the cloud platform using the ROS system and TurtleBot2 robot platform. We set a Kinect driving node on local TurtleBot2 robot for collecting RGB images and the depth images, and publishing on the related ROS topics. We set a 3D scene reconstruction node on the virtual machine of the cloud platform, which subscribes to RGB image topic and depth image topic of Kinect node to compute using 3D scene reconstruction implementation method, so as to upload the intensive computing to the cloud platform.Thirdly, we design and implement the architecture of 3D scene reconstruction using multi robots based on the cloud platform. By designing some functional topics, nodes and modifying launch file on virtual machine of the cloud platform and local robots, this paper implement the structure that a virtual machine of the cloud platform corresponding to a robot, which can use multi robots of the cloud platform for 3D scene reconstruction at the same time. This paper designs a map resource management database to manage the 3D map files on the cloud platform, and achieved the map sharing mechanism between robots and between robots with the cloud platform.