为满足现代施工场地对数字化发展的需要,基于已有的建筑施工现场研究,提出基于计算机视觉的施工现场多目标检测方法.通过SODA(Simple Ocean Data Assimilation)数据集在检测程序中训练,利用学习率改进算法优化训练模型,在多次训练的模型中筛选比对得到最优训练模型;基于无人机自动飞航拍摄的实际施工场地,利用最优的训练模型在不同的施工阶段检测影响施工进度与施工安全的目标,获取检测数据;最后通过创建轻量化的进度安全管理平台,将检测数据创建并导入数据库,在Web平台上实现数据可视化.结果表明:在无人机飞航视角下的目标检测,能够有效统计施工目标与安全情况;与人工方式相比,基于计算机视觉的检测更具理性且效率更高,对于施工现场智能化进度监测与安全管控具备一定的参考价值.
Research on Construction Site Inspection Method Based on Computer Vision
In order to meet the needs of modern construction sites for the development of digitalization,based on the existing research on construction sites,this paper proposes a multi-target detection meth-od for construction sites based on computer vision.The SODA dataset is trained in the detection pro-gram,the learning rate improvement algorithm is used to optimize the training model and the optimal training model is obtained by filtering and comparing the models trained many times.Based on the ac-tual construction site photographed by the UAV,the optimal training model is used to detect the tar-gets that affect the construction progress and construction safety in different construction phases and to obtain the detection data.Finally,by creating the lightweight progress and safety management platform,the detection data is created and imported into the database,and the data is realized on the Web platform.Finally,by creating a lightweight progress and safety management platform,the de-tection data are created and imported into the database and the data are visualized on the Web plat-form.The results show that the target detection under the viewpoint of UAV flight can effectively count the construction targets and safety,and the computer vision-based detection is more rational and efficient compared with the manual method,which is of certain reference value for the intelligent pro-gress monitoring and safety control in construction sites.
万方数据
维普
