Interior design assistant algorithm based on indoor scene analysis

Abstract

The scene analysis algorithm in interior design is widely used in computer vision. To achieve superior interior design outcomes, it is essential to accurately identify and locate indoor objects and structures. However, the common algorithms currently rely too much on color images and manual annotation. Accordingly, the objective of the research is to enhance the interior scene analysis algorithm in interior design, thereby optimizing its performance in the domain of computer vision. In light of the shortcomings of existing algorithms that rely excessively on color images and manually labeled data, this paper employs a dual feature encoder to conduct a comprehensive mining of deep image features, thereby markedly enhancing the precision of semantic segmentation. Then, the accuracy of indoor scene analysis is further improved by integrating the texture features of color images into the modal knowledge distillation of depth images. In addition, to reduce the dependence on manually labeled data, an unsupervised cooperative segmentation algorithm is proposed, which realizes automatic image semantic segmentation through the segmentation process from superpixel to block and then to object. The experimental results showed that the proposed algorithm based on modal knowledge distillation had an average accuracy of 48.29 % in the four types of output. The FIoU value of the unsupervised image cooperative segmentation algorithm reached 66.20, which is superior to the existing algorithms and can better match the real indoor scene. The proposed indoor scene analysis algorithm using color images as privileged information significantly improves the accuracy of indoor scene analysis and reduces reliance on manually annotated data. Moreover, the research algorithm effectively identifies indoor objects, protects personal privacy, and provides a better solution for indoor object analysis.