Accurate Modeling Method of Earthquake Disaster AR Scene Constrained by Structural Semantic Features

Objectives: Earthquake disaster has become one of the natural disasters that pose a great threat to human society because of its suddenness, destructiveness and wide range of impact. Such disasters are usually accompanied by large-scale building damage, loss of life and property, and complex secondary disasters. Therefore, it puts forward high requirements for the efficiency and accuracy of emergency rescue. In this context, Augmented Reality (AR) technology can provide more intuitive decision-making support for rescue personnel by integrating virtual information into the real scene. Therefore, the efficiency and accuracy of rescue can be significantly improved. However, the existing AR scene modeling technology faces challenges in the special environment of earthquake disaster. The main issues lie in the inaccuracy of feature extraction and insufficient precision of the virtual-real fusion modeling, which makes it difficult to achieve the expected accuracy of AR scene modeling in complex earthquake disaster sites. Methods: An accurate modelling method for earthquake disaster augmented reality scenes with structural semantic feature constraints is proposed. Firstly, the characteristics of earthquake disaster scenes are analyzed, and the semantic feature library of earthquake disaster scene structure is constructed. Secondly, the virtual-real feature extraction of earthquake disasters with structural semantic constraints is proposed to improve the accuracy of virtual-real image feature extraction. Then, based on the results of virtual-real feature extraction, the virtual-real modelling of the earthquake disaster augmented reality scene is carried out. Finally, the damaged building is selected as the experimental case area for experimental analysis. Results: Experimental results indicate that the proposed method achieves an F1-score of 90% for earthquake disaster feature extraction based on structural semantic constraints. Additionally, the virtual-real fusion error in the AR scene modelling results is reduced to approximately 20% of the coarse registration error after alignment. Conclusions: The findings demonstrate that the proposed method effectively extracts and matches virtual-real features, leading to accurate AR scene modelling for earthquake disaster scenarios. At the same time, the proposed method has important application potential for improving the accuracy and reliability of AR applied to disaster emergency response, and can assist onsite rescue personnels to obtain more accurate disaster information.