Research On 3D Object Recognition Based On X~2BOT Robot

With the rapid development of the theory and technology of intelligent robot,computer vision has become the key technology of intelligent robot research.Three-dimensional object recognition is one of the basic tasks of computer vision.As far as the overall operation of intelligent robots is concerned,the introduction of computer vision enables robots to identify objects humanely and classify them,and the processor can control the limbs of the machines to perform subsequent actions.Therefore,three-dimensional object recognition based on robot manipulation and humancomputer interaction is a hot research topic in the neighborhood of machine vision.This paper focuses on the real-time recognition technology of indoor threedimensional objects based on Kinect sensors.The RGB-D point cloud is generated by the color information and depth information that collected by Kinect sensors in the indoor environment,and the objects to be recognized in the scene point cloud are identified by the proposed method of fast local feature matching of the scene point cloud model,which can be divided into four steps: Kinect camera calibration,scene point cloud acquisition,complex environment point cloud segmentation and threedimensional object recognition.The research contents are as follows.1.Camera calibration: Using Zhang Zhengyou's calibration method,the Kinect camera is calibrated with a checkerboard to obtain the internal and external parameter matrices of the color camera and the depth camera,and the mapping matrix relationship between the color camera and the depth camera is derived;2.Acquisition of scene point cloud: the color scene information and depth scene information obtained by Kinect camera are filtered and denoised by corresponding depth images,and then the color point cloud of the scene is obtained according to the mapping matrix of color images and depth images;3.Cluttered Point Segmentation: In the indoor 3D object recognition algorithm,it is necessary to segment the complex environment point cloud on the indoor desktop to obtain the 3D point cloud of a single object.By extracting and matching surface normal boundary curves in the point cloud to be segmented,3D bilateral symmetry in the scene is detected,symmetry hypothesis set is established,and complex point cloud segmentation in indoor living environment is completed based on symmetry point attribute judgment;4.3D object recognition: The key points are extracted from the segmented single object point cloud,and the B-SHOT descriptor operator is calculated to establish an indoor object point cloud model library.The KNN algorithm is used to search and match the scene point cloud,and the 3D Hough voting algorithm and geometric consistency algorithm are used to match and filter to realize accurate recognition of three-dimensional objects.The innovation of this paper is to propose a new point cloud segmentation algorithm,improve the point cloud feature descriptor and build a real-time recognition system for indoor 3D objects.The experimental results prove the recognition accuracy and real-time performance of the algorithm.