智慧工地中低分辨率的安全帽状态识别

Low-resolution safety helmet state recognition and its application in smart construction sites

在工地施工过程中,安全帽正确佩戴在保障工人人身安全方面起到至关重要的作用,利用无人机巡检的方式可对工人安全帽佩戴状态进行监督。针对拍摄图像中目标头部的分辨率过低,存在漏检的问题,设计了一种基于深度学习的低分辨率的安全帽状态识别算法,首先通过目标检测网络获取图像中的人体区域,根据先验信息获取头部位置图像,然后使用超分辨率重建技术,对头部图像进行像素恢复,将头部图像输入到目标分类网络中进行判别,最终得到了88.28%的识别正确率。结果表明,该算法可以有效改善漏检问题,符合工地对工人安全帽佩戴识别的要求。

To wear safety helmets correctly plays a critical role in safeguarding personal safety of workers on the construction site.Use of the unmanned aerial vehicle(UAV)patrol shooting can supervise workers’wearing status of safety helmets.In response to the extremely low-resolution rate of head images and the omission of inspection,a low-resolution safety helmet status recognition algorithm is designed based on deep learning.First of all,the target inspection network is used to acquire the human body area in the images.Then,the head position images can be obtained according to prior information.Next,the super-resolution reconstruction technology is used to pursue pixel recovery of head images.The head images are then input the target classification network for recognition,and the correct recognition rate is as high as 88.28%.Results show that this algorithm can effectively improve the problem of inspection omission and meet the construction site’s requirement of workers’safety helmet wearing recognition.