Research And Implementation Of 3D Recognition Of Engine Parts Based On Mixed Reality Technology

Due to the complex structure of the aero engine body,it is difficult to disassemble and assemble,and is not easy to maintain.Novice operators are not familiar with various types of engine parts.In the field of aero engine replacement,there are problems such as difficulty in training and high practical cost.Therefore,they often use The target detection method classifies and detects the types of parts.However,in the real world,the parts are stacked in a mess and are prone to occlusion.The traditional two-dimensional recognition method of a single target will not be able to extract enough feature information,and it is easy to have wrong detection results,which cannot meet the needs.At the same time,the traditional replacement The training process is based on paper or electronic documents,and the training efficiency is low.Based on the above problems and research difficulties,this paper proposes to add 3D target detection that can use depth information to the field of engine replacement for research,and combine the relevant knowledge of neural networks to obtain the classification,spatial position and pose information of complex engine parts;The combination of mixed reality technology can intuitively feel the virtual inspection information of each part by holographic projection to the eyes,and interact with the virtual part model in real time in a new interactive way,so that novice operators can get a more intuitive experience.Therefore,this article focuses on the related technologies of 3D target detection and mixed reality,and studies and implements the following contents:(1)Unlike two-dimensional detection that only recognizes a single target,aero-engine parts are often complex and scattered with each other,and the use of two-dimensional target detection is likely to fail to adapt to a messy environment.This paper uses a three-dimensional target detection algorithm based on deep learning to achieve accurate detection of engine parts.At the same time,in order to improve the recognition accuracy of parts and the robustness in complex environments,this paper uses spatial context-aware mixed reality methods to generate parts data sets for training and testing.(2)Aiming at the unique spatial mapping function of the new generation of Holo Lens2 mixed reality helmets,using RGB-D camera-assisted tracking and registration technology to complete the registration of the virtual engine part model in the mixed reality scene,and design a new type of gesture and voice command Way to realize real-time interaction with the part model.(3)On the basis of the previous research points,this paper innovatively proposes the key technology of integrating mixed reality and three-dimensional target detection to solve the problems in the field of engine replacement.By combining the 3D part inspection model with the Holo Lens2 mixed reality helmet,a part recognition interactive system based on Holo Lens2 is realized.The system can identify and locate engine parts in a complex environment,and combine with the SLAM system to complete the continuous and stable tracking of the parts,realize the continuous positioning of the detected parts in the space,and complete the mixed reality effect.The experimental results show that the method in this paper combines deep learning-based three-dimensional target detection with key technologies of mixed reality,and designs simple and natural interaction methods,which can realize the identification and detection of engine parts with high accuracy.The innovation can provide a new visual interactive experience for parts inspection and reissue training and learning in the industrial field.