Research On Assembly Training Software Platform Based On Augmented Reality Technology

At present,product assembly in the industrial field mainly relies on manual methods,for complex electromechanical products assembly requires experience and skills,the operator can master only after manual training,this traditional training guidance assembly method is inefficient,low standardization,high proficiency requirements and easy to cause errors.Augmented reality technology is a technology that can integrate virtual assembly tips into the real world and can provide efficient guidance or training to operators.This thesis investigates a combination of natural feature point-based and edge contour feature-based methods to develop an augmented reality-based assembly training software platform.Firstly,to improve the feature point detection,description and matching method of traditional ORB algorithm for the problem of low accuracy and poor stability of natural feature point based 3D registration method in augmented reality technology.Secondly,a contour feature-based method is used to identify the parts and prompt the operator with text messages for the assembly steps.Finally,the guidance process of assembly is demonstrated with a vacuuming robot as the object.The details of the research are as follows.1.Improving ORB algorithm in natural feature point based tracking registration.By analyzing the principles and drawbacks of the traditional ORB algorithm,the feature point detection,description and matching methods of the ORB algorithm are improved.Firstly,we propose to use Harris operator to solve the problem that feature points detected by oFAST operator are easy to produce from clustering phenomenon.Then the histogram method is used to improve the extraction mode of the principal direction,which solves the problem that the principal direction changes with the rotation of the neighborhood in the gray-scale prime center method.Finally,to address the problem that the traditional ORB algorithm is prone to error when violent matching or Hamming distance matching,FLANN-based matching and RANSAC-based algorithm are used to screen out the mismatch.And the improved algorithm is verified in the augmented reality system to improve the accuracy while satisfying the real-time performance.2.For the needs of part recognition and assembly step recognition arising in the assembly process,the edge contour feature-based part recognition method is proposed,which can effectively solve the problem that the natural feature point-based method cannot detect the parts with insufficient texture.Preprocessing,edge detection,contour finding and calculation of Hu invariant moment of the detected image are performed,and the search strategy in the traditional invariant moment matching method is improved to traverse the image with the center-of-mass coordinates of the search subgraph as the reference,and the parts are recognized based on the improved matching method.3.Using the vacuum robot as the assembly object,we design and render a virtual model,build an assembly training software platform based on augmented reality technology,and demonstrate the part recognition module and auxiliary assembly module in the system to realize the 3D visualization,virtual-reality combination and part recognition function of the assembly training process.The feasibility and effectiveness of the method combining natural feature point tracking-based registration with edge contour feature-based part recognition are verified.