Based On Mobile Augmented Reality Assist Vehicle Maintenance Methods Research

With the increasing of automobile ownership,automobile maintenance industry has formed a strong industrial system.Because most of the owners lack of knowledge about the car,they need to ship the car to a repair shop or 4S shop for maintenance when they encounter faults in the driving process.At the same time,with the continuous development of automobile manufacturing industry,the structure and principle of automobile parts are increasingly complex,which undoubtedly increases the cognitive load of maintenance personnel.In addition,most of the current autonomous vehicle maintenance methods lack reasonable maintenance process information and interaction modes,so the maintenance efficiency is often low for those users who are not familiar with the maintenance steps.Therefore,this paper proposes an auxiliary vehicle maintenance scheme based on mobile augmented reality technology.By superimposing virtual auxiliary information into the real maintenance environment and visualizing it,it can not only improve the maintenance efficiency of maintenance personnel,but also greatly reduce the maintenance cost of users.The specific research content of this paper mainly includes the following three aspects:(1)A maintenance target recognition algorithm based on improved YOLOv5 s is proposed.In view of the disadvantages of the traditional recognition algorithm of maintenance parts based on image or point cloud features,such as poor robustness,large computation and high requirements on hardware equipment,this paper selects YOLOv5 s lightweight neural network model,which is currently excellent in model reasoning speed and detection performance,as the recognition algorithm of maintenance targets.Aiming at the problem of low accuracy of detection due to the mixed target and complex background of automobile parts in the maintenance scene,SENet attention mechanism was introduced into the network to enhance the feature representation ability of the target in the complex background.In addition,K-Means algorithm was used to re-cluster the self-made data set of auto parts to obtain the optimal size of the initial anchor frame.Finally,through model training and testing,the experimental results show that the improved network recognition accuracy is greatly improved,but also meet the real-time requirements of the system.(2)An algorithm for monocular visual inertial SLAM for augmented reality tracking registration is studied.In the vision front end,aiming at the problem of uneven ORB feature point extraction,a uniform ORB feature extraction algorithm based on dynamic threshold and hierarchical quadtree is proposed,and the IMU constraint between image frames is processed by pre-integration method.Aiming at the problems of incomplete initialization parameters and slow initialization speed in current visual inertial initialization,a joint initialization method of visual inertial navigation based on maximum posterior estimation was proposed,which can achieve rapid initialization within 5 seconds and initialize all parameters at one time.In the SLAM back-end,a nonlinear optimization method based on sliding window is used to estimate the pose state of the system,and a key frame screening mechanism and marginalization operation are established to keep the computational amount of the system constant.In addition,a loopback detection method based on Bo W(word bag model)is used for loopback detection,and loopback constraints are established to optimize the global pose of the system.Finally,the experimental analysis is carried out in Eu Roc standard data set and compared with the classical VINS algorithm.The experimental results show that the proposed method has higher pose tracking accuracy.(3)A smart phone based on augmented reality assisted vehicle maintenance application is developed.The maintenance target recognition algorithm proposed in this paper is fused with monocular visual inertial SLAM algorithm to realize the real-time tracking of the maintenance target by THE SLAM system.The fusion algorithm is imported into the Unity3 D rendering engine in the form of dynamic link library,and the pose result calculated by the SLAM system is used as the input of the virtual plug-in camera in Unity3 D.The integration of real world and virtual world is completed in virtual camera.Finally,an augmented reality assisted vehicle maintenance application was developed on Unity3 D,which was packaged and deployed on smart phones to realize rendering and registration of virtual maintenance guidance information and animation.