Safety helmet detection algorithm based on improved YOLOv5 model

Abstract

The invention relates to a safety helmet detection algorithm based on an improved YOLOv5 model, and the algorithm introduces an attention mechanism thought, combines the attention mechanism thought with a feature point description mode, captures more effective information, and improves the robustness of descriptors. An anchor frame is generated on three different scales in a clustering mode through a k-means + + algorithm to serve as an initial frame of a model, an ARM is added into the model, then the ARM and a YOLOv5 backbone network are linked through a BFF, dynamic prediction and better target locking can be carried out by adding an FSM and a DRH, a final candidate frame is obtained through weighted frame fusion through a WBF algorithm, and the target detection precision is improved. Adopting an acceleration model to improve the detection performance of the model; according to the method, the precision of small-size target detection is improved, the probability of missing detection and false detection is reduced, the requirement of real-time detection is met, the precision can be ensured under the conditions that the picture data background is complex and the like, and the method is closer to the actual application scene.

Description

technical field [0001] The invention relates to the combined application field of electronic information and construction engineering, in particular to a safety helmet detection algorithm based on an improved YOLOv5 model. Background technique [0002] Hough's method has been used to identify hard hats as early as 2004. However, what this method obtains is the high-definition image information of the close-range indoor, obviously this is not suitable for the remote monitoring of the construction site. Another one, represented by Cai Limei, uses information such as the direction of the external properties of the helmet to simulate the helmet, which is suitable for helmet detection in a more complex environment such as collecting mine information, but it needs to match the accurate information features from the outside. The above detection methods are mainly researched on the basis of the arc shape characteristics of the helmet. [0003] At present, there are also comprehens...

AI Technical Summary

Problems solved by technology

With the deepening of research, neural network structures with multiple layers and complex structures (LeNet5, AlexNet) have been proposed, which also means that the function of target detection is becoming more and more perfect. However, in the real construction environment, helmet detection The accuracy of small-sized target detection is low

Images