3D Indoor Room Layout Estimation Using a Deep Learning–Driven ConvBi-LSTM Model for Marine and Maritime Applications

Abstract

Estimating room layout has become increasingly important in recent years because of its wide range of applications,
particularly in marine environments such as ship interiors, offshore platforms, and maritime facilities. However, this
task is quite challenging due to factors like clutter, occlusion, and varying lighting conditions in room images, which are commonly observed in marine and maritime operational settings. Accurately predicting the 3D layout of a room from just a single 2D image is crucial for applications including marine safety systems, autonomous inspection of vessels, and smart ship design. Existing techniques attempt to address this problem but often rely on a limited set of features. For better prediction accuracy, models need to process a larger number of features. To address this, we propose a robust framework for 3D marine room layout estimation in indoor environments. The process begins with pre-processing the input image, after which our model generates both edge maps and semantic labels. For layout prediction, the framework employs a Deep ConvBi-LSTM model. A score function is then optimized using the Remora Optimization Algorithm (ROA) to select the best 2D layout from the candidate set. Finally, the 3D layout is reconstructed from the chosen 2D layout using layout coordinates and camera orientation. Experimental results demonstrate that the proposed model performs effectively and delivers strong results in marine and maritime indoor layout estimation scenarios.