Intelligent Building Detection And Registration Method In Mobile Augmented Reality

Augmented reality technology is a technology that can combine the virtual scene simulated by computer with the real world.It has broad prospects in the fields of data visualization and human-computer interaction.With the rapid development of mobile computing devices and mobile Internet,mobile augmented reality technology has gradually attracted the attention of academia and industry,especially in outdoor largescale scene applications.The research of image recognition and registration for buildings has high practical value and can provide useful approaches for city guide,navigation,and wayfinding.Registration is the core of mobile augmented reality technology.In the outdoor scene,the early method based on satellite positioning and sensor is susceptible to system error and positioning accuracy,resulting in poor registration effect.The method based on vision-based markers is difficult to broadly use in outdoor scenes.Nowadays,more methods use visual natural feature extraction to achieve object recognition,tracking,and registration in outdoor scenes.These methods,however,are susceptible to outdoor conditions such as illumination changes and dynamic occlusion,and they also require the construction of image feature matching libraries for specific objects.Furthermore,the computing tasks of these methods are usually done on the server side,which limited their universality,flexibility,and real-time performance.Based on the aforementioned problems,this thesis firstly designed a lightweight deep learning object detection model named Squeeze Net SSD,then combined it with a deep transfer learning strategy to achieve intelligent detection of buildings,which effectively avoided the limitations of traditional image feature matching methods,leading to the improvement of its universality.In addition,this model is mobile-oriented,which means it can make full use of mobile computing resources,avoid network delay and dependence,and guarantee real-time performance.After that,based on the model,this thesis designed a mobile augmented reality hybrid registration method for the buildings in outdoor scenes,which included building information matching through orientation,multi-object parallel tracking and error recovery through kernel correlation filter algorithm,and the conversion between the screen coordinate system and the camera coordinate system based on the visual detection results,satellite positioning,and sensors.The development and test of the application examples showed that the building intelligent detection model and registration method designed in this thesis can achieve precise building detection,and obtain visually accurate virtual-real fusion,leading to the improved performance of the augmented reality display and interaction on the buildings in outdoor scenes.This thesis made useful explorations in securing and improving the universality,flexibility,and real-time performance of mobile augmented reality,and it also provided new ideas and technical support for the broader application of mobile augmented reality in outdoor scenes in the future.