Research On Safety Helmet Wearing Detection Algorithm Based On Deep Learning

Object detection is one of the hot topics in the field of computer vision,and it is widely used in intelligent driving and smart cities.The application of object detection technology to safety production has always been a cross-field combination advocated by the state in recent years.It is of great significance to detect the wearing of safety protective gear.Traditional methods use image processing and machine learning to detect wearing helmets,which have low detection accuracy and slow detection speed,which cannot meet the needs of use.Safety helmet wearing detection uses deep learning technology to greatly improve the detection accuracy,but the existing network model still has the phenomenon of false detection and missed detection in complex work scenarios.Based on the YOLOv3 network model,this paper studies the helmet wearing detection algorithm in complex work scenarios.The specific research is as follows:First,point at the missed detection and misclassification caused by the occlusion of dense crowds in the current helmet wearing detection algorithm,this paper proposes a multi-component classifier.The classifier is designed according to factors such as the aspect ratio of the person detection frame,the angle between the helmet-person center point vector and the vertical direction,and the spatial distance ratio.The obtained classifier model meets the expected effect of the experiment and effectively improves the classification accuracy.Using the Soft-NMS algorithm to improve the original NMS algorithm in YOLOv3 can effectively reduce the missed object detection caused by the occlusion objects with same label.Second,address the problems of low object detection accuracy and serious missed detection in complex work scenarios and small object scenarios.This paper proposes a helmet wearing detection algorithm that combines inverted residual neural network and attention mechanism.The CBAM attention mechanism module is embedded in the feature extraction network Dark Net53 of YOLOv3 to calculate the weights of the channel domain and spatial domain of the feature map,and multiply the weight with the input feature map to obtain a feature map that combines the weights of the attention mechanism.Increase the object feature channel weight,suppress the background information channel weight,and improve the accuracy of object detection.Using deep separable convolution to replace the traditional convolution method in the Dark Net53 residual network,the inverted residual network is obtained,which effectively reduces the amount of calculation in the feature extraction process and improves the speed of model detection.Finally,through a large number of experimental verifications,The experimental results show that the helmet wearing detection algorithm proposed in this paper,which combines the anti-residual neural network,the attention mechanism and the multicomponent classifier,is scientifically fused and designed,and the average detection accuracy rate reaches 93.7%.Compared with the existing algorithm using the YOLOv3 model,the detection speed is increased by 21%,and the detection accuracy is 5.4%higher;the detection accuracy is 0.4% and 1.2% higher than the latest methods using the YOLOv4 and YOLOv5 models respectively.It can meet the practical application requirements of helmet detection in most work scenarios.