An Improved Fuzzy Decision and Geometric Inference-Based Indoor Layout Estimation Model From RGB-D Images

Indoor layout estimation is a critical aspect of indoor scene understanding, aiming to recover and reconstruct the geometric structure information of indoor spaces by analyzing images or depth data. The indoor layout estimation is a challenging task due to the complexity of the indoor environment, including the unstructured geometric construction and complex illumination conditions. To address these issues, an improved fuzzy decision and geometric inference-based indoor layout estimation model from red, green, blue, and depth (RGB-D) images is proposed in this article. In the proposed model, to address the challenges of fixed plane detection thresholds, missing wall planes, and depth data loss due to transparent or reflective materials, three main improved modules are presented, namely, the fuzzy decision-based threshold adjustment (FDTA) module, the region growing-based wall supplement (RGWS) module, and the geometric inference-based depth completion (GIDC) module. The FDTA module is used to optimize the plane detection results based on the initial plane detected of the wall and floor to improve the accuracy and robustness of layout estimation. Then, the RGWS module supplements missing wall planes in the preliminary detection results, while the GIDC module completes missing depth information due to transparency or reflectivity in the input images. Experimental results show that the proposed method significantly improves the accuracy and robustness of indoor layout estimation, providing a reliable and efficient solution for complex indoor scenes.