基于DeepFlux算法的建筑施工脚手架间距检测

Spacing detection of construction scaffolding based on DeepFlux algorithm

在脚手架坍塌事故仍有发生的背景下,为了避免传统的人工脚手架测量方法低效和高危的缺点,因此采用计算机视觉对脚手架进行安全规范检测的方式。首先,采用DeepFlux算法提取脚手架图片的骨架信息,针对提取效果以及精度不能满足实际需求的问题,将DeepFlux算法中的VGG16特征提取网络替换为InceptionV3网络,有效地提高了骨架提取精度。其次,根据提取到的骨架信息,提出一种交点检测算法计算脚手架交点信息。最后,根据交点信息计算得到脚手架杆间像素间距,再采用标靶法换算成实际间距。测试实验结果表明,在对脚手架进行检测的任务中,计算脚手架参数的平均误差在5%左右,满足脚手架检测的准确性,能够做到代替人工测量实现脚手架的安全规范检测。

In the context of occasional scaffold collapse accidents,in order to avoid the inefficiency and high-risk drawbacks of traditional manual scaffold measurement methods,computer vision is used to detect scaffolds in a safe and standardized way.First,the DeepFlux algorithm is used to extract the skeleton information of scaffold pictures,and for the problem that the extraction effect as well as the accuracy cannot meet the actual demand,the VGG16 feature extraction network in the DeepFlux algorithm is replaced by the InceptionV3 network,which effectively improves the skeleton extraction accuracy.Secondly,according to the extracted skeleton information,an intersection detection algorithm is proposed to calculate the scaffold intersection information.Finally,the pixel spacing between scaffold poles is calculated based on the intersection point information,and converted to the actual spacing using the target marking method.The test experiment results show that the average error of calculating scaffold parameters in the task of testing scaffolds is about 5%,which satisfies the accuracy of scaffold testing and can achieve the safety specification testing of scaffolds instead of manual measurement.