A safety helmet wearing detection method, device and medium

Abstract

The invention discloses a safety helmet wearing detection method, device and medium. In order to solve the common problems of low detection accuracy and poor robustness in existing helmet detection algorithms, this algorithm proposes a novel helmet wearing detection method based on the SSD algorithm based on multi-scale perception and feature adaptive fusion. The algorithm deconvolves high-level features and adaptively fuses them with low-level features, rather than simply adding or multiplying them in equal proportions, thereby generating new and more efficient feature maps. In addition, the algorithm also introduces a multi-scale perception module to the new high-level feature map to improve the robustness of the network to target scale changes. Finally, the algorithm also designs an effective anchor box allocation strategy on the output feature pyramid, which can adaptively adjust the scale distribution of each layer of anchor boxes according to the size of the output feature map, which is beneficial to detect small features on the low-level feature map. The experimental results show that the algorithm achieves a high detection level on the hard hat detection dataset GDUT‑HWD.

Description

technical field [0001] The invention relates to a safety helmet wearing detection method, device and medium, belonging to the technical field of computer vision. Background technique [0002] In construction sites, mining areas, electric power and chemical operation areas, incidents of falling objects injuring people often occur. Therefore, anyone entering the above-mentioned relevant work areas is required to wear safety helmets to ensure the safety of personnel. But from now on, there are still many problems in the safety management of these industries. First of all, from the perspective of construction personnel, it is difficult to fully cover safety education, and there are always construction personnel who are lucky and have low risk awareness, and do not wear safety helmets as required. Secondly, from the perspective of supervision, most enterprises and supervision departments still rely on special personnel to monitor whether construction workers wear safety helmets,...

AI Technical Summary

Problems solved by technology

But from now on, there are still many problems in the safety management of these industries

First of all, from the perspective of construction personnel, it is difficult to fully cover safety education, and there are always construction personnel who are lucky and not strong in risk awareness, and do not wear safety helmets as required

Secondly, from the perspective of supervision, most enterprises and supervision departments still rely on specialized personnel to monitor whether construction workers wear safety helmets, or set up warning signs on dangerous parts of the project to implement supervision, which has low supervision efficiency and timeliness. Therefore, these traditional means are increasingly unable to meet the needs of existing security management

[0003] The automatic detection of whether to wear a helmet is a kind of target detection problem, but there is a problem in the field of target detection, that is, the detection effect on small targets and targets with large changes in target scale is poor, and the detection of wearing a helmet is somewhat It is difficult to detect objects belonging to small and medium scales.

Images