Automated Window Opening Segmentation and Levelness Measurement

Conventional wall opening inspections, including levelness checks, rely on manual surveying, which is time-consuming and labor-intensive. Recent research has explored automated image-based methods, with image segmentation playing a key role. Red-green-blue-depth (RGB-D) images, which include depth information, have been used to improve segmentation accuracy. However, these efforts are limited in using depth maps in measurement tasks. To address this gap, this paper proposes a deep learning-based method for detecting wall openings and measuring their levelness using RGB-D images. The approach includes (1) segmenting window openings in indoor scenes using a deep learning-based model, and (2) identifying window corners and computing levelness based on depth values and a horizontal baseline. The method was evaluated on real-world indoor scenes, achieving 85.7% intersection over union (IoU) for segmentation, 70.52 mm mean absolute error (MAE), and 3.52% mean relative error (MRE) for measurement.