Research On Textureless Target Recognition Technology Based On Edge Sector Distribution Features

In order to meet the flexible production requirements of the factory,modern factories are gradually increasing the use of intelligent sensing robots.Identifying the target object is one of the core applications of the intellisense robot.For the implementation problem of target object recognition,it is a solution to extract features from industrial camera images to match and determine the category.For the textureless target object,the existing characterization algorithm matching accuracy is low:SIFT,SURF and other features have a large number of mismatches in the recognition application of textureless targets,and the calculation speed is slow;ORB,BRIEF and other binary features Poor force and insufficient robustness.Aiming at the above deficiencies,this paper proposes an edge sector distribution feature around the recognition problem of textureless targets,and uses this feature to realize the identification and sorting of textureless objects.The first chapter introduces the research status of image segmentation and target recognition at home and abroad,classifies and summarizes various related algorithms,and expounds the research significance of the problem of non-texture target recognition.Then the research content and the full-text framework of this paper are introduced.The second chapter preprocesses the image of the textureless target,mainly including image enhancement and region of interest extraction,in preparation for feature point extraction,matching and target recognition in dynamic scenes.The non-subsampled contourlet transform algorithm is used to enhance the contour of the target.The histogram bimodal method is used to mark the foreground image of the target image.Based on the marker,the watershed algorithm is applied to segment the region of interest of the target image.In the third chapter,the feature points are extracted,the descriptors are constructed and matched,and the robustness of the proposed features is analyzed and tested.The Harris corner points are extracted and selected as feature points.The straight line segments and arc segments at the edge of the image are reconstructed to construct feature point descriptors.The feature was then subjected to scale analysis and robustness evaluation experiments.The fourth chapter proposes a recognition and sorting method for textureless objects based on the distribution characteristics of edge sectors and the problem of identification and sorting of textureless objects.This chapter mainly describes the image processing process of identifying the sorting method.The DBSCAN algorithm is used to cluster the feature points of multiple targets.After identifying the multi-target of the initial frame and marking the initial position,the particle filter algorithm is used to track the multi-target.Industrial robots perform sorting tasks.Finally,the effectiveness of the proposed recognition and sorting method is verified by comparing the experimental results of various algorithms.The fifth chapter builds a software and hardware platform for the textureless target recognition system.The hardware composition,selection,connection between modules and the hand-eye calibration process of industrial cameras are introduced.The user interface of the texture-free target recognition system is constructed.The intermediate process and the resulting images and data are displayed,and the images are integrated.Processing modules such as processing,industrial robot kinematics solution,system communication,etc.The sixth chapter summarizes the advantages and disadvantages of the full text,and looks forward to the follow-up research.