基于机器学习方法的安全帽佩戴行为检测

Detection on wearing behavior of safety helmet based on machine learning method

施工现场作业人员是否佩戴安全帽主要依靠人工检查,存在监管效率低、时效性差等问题,为了实时自动监管作业人员是否佩戴安全帽,提出1种基于机器学习的安全帽佩戴行为检测方法。首先利用深度学习YOLOv3算法检测出现场视频中的施工人员脸部位置,根据安全帽与人脸的关系估算出安全帽潜在区域;然后对安全帽潜在区域图像进行增强处理,使用HOG(方向梯度直方图)提取样本的特征向量;再利用SVM(机器学习的支持向量机)分类器对脸部上方是否有安全帽进行判断,进而实现对施工人员安全帽佩戴行为的实时检测与预警。以某高铁站施工现场为例进行验证,研究数据表明在施工通道和塔吊作业区域,该方法可实时有效检测出工人未佩戴安全帽的行为,识别率达90%。

Whether the construction site operators wear safety hats mainly depends on manual inspection,which has some problems,such as low supervision efficiency and poor timeliness.In order to supervise whether the workers wear the safety helmet in real-time and automatically,a detection method for the wearing behavior of safety helmet based on the machine learning was put forward.Firstly,the face position of worker in the field video was detected by using the deep learning YOLOv3 algorithm,and the potential area of safety hamlet was estimated according to the relationship between safety hamlet and personnel face.Secondly,the enhancement processing was carried out on the image of potential area of safety hamlet,and the characteristic vector of sample was extracted by using the histogram of oriented gradients(HOG).Finally,the support vector machine of machine learning(SVM)classifier was applied to judge whether the safety hamlet was above the face,thus realize the real-time detection and early-warning on the wearing behavior of safety helmet of the workers.The verification was carried out by taking the construction site of a certain high-speed railway station as an example,and the results showed that in the areas of construction passage and tower crane operation,the behavior of without wearing the safety hamlet of the workers could be detected in real-time and effectively by using this method,and the recognition rate reached 90%.