Automatic Drywall Analysis for Progress Tracking and Quality Control in Construction

TL;DR

This work tackles automated drywall progress tracking and quality control by introducing an image-based drywall analysis pipeline that fuses a specialized instance segmentation model with a geometry-driven analysis module. The instance segmentation component uses a modified Mask R-CNN with a ConvNeXt V2 backbone, augmented by tailored anchor ratios and higher-resolution masks ($56\times56$), and is trained with a targeted data augmentation strategy to overcome limited data. The drywall analysis module performs geometry optimization, wall-segment clustering via vanishing points, and perspective correction to yield orthographic projections, enabling precise segment-level progress estimation and quality checks. On a 176-image drywall dataset, the approach achieves substantial segmentation gains (e.g., bbox mAP up to $0.64$ and mask mAP up to $0.60$) and demonstrates practical, real-world applicability for automated, on-site progress tracking and compliance verification.

Abstract

Digitalization in the construction industry has become essential, enabling centralized, easy access to all relevant information of a building. Automated systems can facilitate the timely and resource-efficient documentation of changes, which is crucial for key processes such as progress tracking and quality control. This paper presents a method for image-based automated drywall analysis enabling construction progress and quality assessment through on-site camera systems. Our proposed solution integrates a deep learning-based instance segmentation model to detect and classify various drywall elements with an analysis module to cluster individual wall segments, estimate camera perspective distortions, and apply the corresponding corrections. This system extracts valuable information from images, enabling more accurate progress tracking and quality assessment on construction sites. Our main contributions include a fully automated pipeline for drywall analysis, improving instance segmentation accuracy through architecture modifications and targeted data augmentation, and a novel algorithm to extract important information from the segmentation results. Our modified model, enhanced with data augmentation, achieves significantly higher accuracy compared to other architectures, offering more detailed and precise information than existing approaches. Combined with the proposed drywall analysis steps, it enables the reliable automation of construction progress and quality assessment.

Figures (9)

• Figure 1: Instance segmentation and drywall analysis pipeline.
• Figure 2: Proposed modification of the Mask R-CNN architecture. We replace the backbone used in the original paper with the ConvNeXt V2 model, introduce additional ratios for computing anchor boxes, and add a block of deconvolutional layers at the end of the instance mask branch to increase the output mask resolution.
• Figure 3: Example of synthetic images. The left image shows augmentation with randomized placement of cropped elements, while the right image demonstrates structured placement.
• Figure 4: Visual comparison of detected metal frames (first row, in red) and wood panels (second row, in blue) by the baseline model and the modified architecture.
• Figure 5: Instance segmentation results using different data augmentation techniques. Column (a) shows the original image, (b) the baseline output, (c) the baseline with data augmentation using COCO backgrounds and random placement, and (d) the baseline with data augmentation using room-interior backgrounds and structured placement. Detected wall panels are shown in pink, wood panels in blue, insulation in orange, and metal frames in red.