School of Mechanical-Electronic and Vehicle Engineering, Beijing University of Civil Engineering and Architecture, Beijing 100044, China
Show less
文章历史+
Received
Published
Revised
2025-10-20
2026-09-18
2026-01-19
Online Date
2026-10-09
PDF (16479K)
摘要
针对建筑门窗传统视觉测量方法精度不足和激光点云扫描数据处理复杂的难题,通过将视觉与激光技术相融合,并利用深度学习优化图像处理流程,基于改进YOLOv11(you only look once version11)提出了一种建筑门窗视觉测量方法。利用自主研发的测量设备采集含有深度信息的多视角图像,通过深度信息与位姿信息完成图像校正以拼接并构建包含待测目标墙面的全景图。为提高门窗测量精度的同时减少计算量,采用YOLOv11-CUA(YOLOv11n-C3k2_UIB-ADown)进行目标检测,并结合LSD(line segment detector)直线特征检测算法实现目标区域测量。改进的YOLOv11引入C3K2_UIB模块来增强多尺度特征提取,提升小型门窗检测能力;ADown模块通过轻量化设计减少计算量,替换下采样卷积以增强关键区域的表征能力,减少低对比度场景的细节丢失。在自建门窗数据集上,门窗检测准确率达97.5%,mAP50为97.2%、参数量仅为2.0 M。相比于YOLOv11n网络,准确率提高了4.7%,mAP50提高了4.8%,参数量减少了0.6 M。最终尺寸平均测量误差为5.63 mm,最大误差控制在10 mm以内,满足建筑测量规范。为智能建造提供了一种高效的自动化测量方案。
Abstract
To address the challenges of limited accuracy of conventional visual measurement methods for interior doors and windows in buildings, as well as the high computational complexity of laser point cloud data processing, a vision-based measurement method was proposed. Visual and laser technologies were integrated, and deep learning models were employed to optimize and streamline the image processing pipeline. Multi-view images with depth information were captured using a self-developed measuring device. Geometric correction and image stitching were performed using depth information and pose data to construct panoramas of walls containing the target doors and windows. To enhance measurement accuracy while reducing computational load, a YOLOv11-CUA (YOLOv11n-C3k2_UIB-ADown) was utilized for object detection, combined with LSD (line segment detector) algorithm to achieve precise measurement within the target regions. The improved model of YOLOv11 includes the introduction of the C3K2_UIB module to strengthen multi-scale feature extraction and improve detection of small doors and windows. The ADown module employs a lightweight design to reduce computational overhead and replaces conventional downsampling convolutions to strengthen feature representation in key regions, thereby reducing information loss in low-contrast scenarios. Evaluated on a dedicated door-window dataset, the system achieves a detection accuracy of 97.5%, a mAP50 of 97.2%, and contains only 2.0 M parameters. Compared with the baseline YOLOv11n, the proposed model shows an improvement of 4.7% in accuracy, 4.8% in mAP50, and a reduction of 0.6 M parameters. The resulting dimensional measurements exhibit an average error of 5.63 mm, with maximum deviations controlled within 10 mm, satisfying relevant building measurement standards. This vision-laser fusion method for measuring interior doors and windows in buildings provides an efficient automated measurement solution for intelligent construction.
LI Pei-xuan, ZHOU Hui-xing, XU Chong-wen, LI Hao-yu, WU Hao.
Visual Measurement of Building Windows and Doors Using Improved YOLOv11-CUA Framework[J].
Science Technology and Engineering, 2026, 26(26): 11406-11416 DOI:10.12404/j.issn.1671-1815.2507317
使用YOLOv11-CUA模型对获取的门窗图进行目标检测,输出门窗的边界框,该步骤提供粗略的门窗定位,为后续直线特征检测提供感兴趣的区域(region of interest,ROI)。相比于传统的Hough变换,LSD无须参数优化且具有线性时间复杂度,能够实时检测有限长度的线段,并在噪声环境下表现出色,更适合建筑门窗的矩形框结构检测,因此,在检测到的边界框内通过LSD直线检测算法进行门窗洞口的直线目标检测,以确认门窗的直线框边界。
GaoYadong, ZhaoLifeng, LiuShibiao, et al. Application of 3D laser scanning technology in the integration of multiple measurements in building construction[J]. Beijing Surveying and Mapping, 2023, 37(3): 371-376.
ZhaoHui, HanZhiwei. Research progress of image feature matching method based on deep learning[J]. Computer Science, 2023, 50(1): 144-150.
[7]
ZolanvariS M I, LaeferD F, NatanziA S. Three-dimensional building façade segmentation and opening area detection from point clouds[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2018, 143: 134-149.
MoYuxiao, XuJingzhong. A method for detecting doors and windows in facade point clouds based on cavity detection and rule constraints[J]. Journal of Lasers, 2025, 46(1): 61-67.
ZhengXiaoyu, ZhouHuixing, WangShun, et al. Measurement of door and window opening dimensions based on laser point calibration and machine vision[J]. Science Technology and Engineering, 2022, 22(15): 6204-6209.
ShenLe, LiGuiqing, XianChuhua, et al. Door and window detection based on indoor 3D point Cloud model[J]. Journal of Computer-Aided Design & Computer Graphics, 2019, 31(9): 1494-1501.
[16]
ChengS, HanY, WangZ, et al. An underwater object recognition system based on improved YOLOv11[J]. Electronics, 2025, 14: 14010201.
[17]
ChenY, ZhangC, ChenB, et al. Accurate leukocyte detection based on deformable-DETR and multi-level feature fusion for aiding diagnosis of blood diseases[J]. Computers in Biology and Medicine, 2024, 170: 107917.
Tursun MaimaitiQiuJianzhuo, LiuJian, et al. Road crack detection algorithm based on improved YOLOv8n[J]. Science Techno-logy and Engineering, 2025, 25(14): 6044-6053.
[20]
HeL, ZhouY, LiuL, et al. Research and application of YOLOv11-based object segmentation in intelligent recognition at construction sites[J]. Buildings, 2024, 14: 3777.
[21]
WangD, TanJ, WangH, et al. SDS-YOLO: an improved vibratory position detection algorithm based on YOLOv11[J]. Measurement, 2025, 244: 116518.
XuXiangyang, YuanShanshan, WangJun, et al. Image stitching method based on global and local features[J]. Journal of Beijing Institute of Technology, 2022, 42(5): 502-510.
[24]
KhanamR, HussainM. YOLOv11: an overview of the key architectural enhancements[J]. arXiv Preprint, arXiv: 2410.17725, 2024.