Interior scene semantic segmentation and painting system design for complex space

Due to the rapid development of the real estate industry, the demand for interior decoration and design is increasing. The application and development prospects of interior coating technology are very considerable. However, due to the complex indoor scene environment, there is mutual occlusion between the target objects, which reduces the quality. In this paper, the depth image is fused with red-green-blue image, and the semantic segmentation model of cross-attention based on fusion depth and the semantic segmentation model of three branches based on coordinate attention are constructed. Simulation experiments show that, in the dataset of NYU-Dv2, the average intersection ratio of the two models constructed in this study is 71.4% and 66.8% in the general scene, and 49.1% and 50.2% in the indoor scene containing many small objects, both of which are superior to other comparative semantic segmentation models. It indicates that the model designed in this study has good semantic segmentation effect, and can help improve the performance of indoor scene coating system.