Hardhat-Wearing Detection Based on a Lightweight Convolutional Neural Network with Multi-Scale Features and a Top-Down Module

Abstract

Construction sites are dangerous due to the complex interaction of workers with equipment, building materials, vehicles, etc. As a kind of protective gear, hardhats are crucial for the safety of people on construction sites. Therefore, it is necessary for administrators to identify the people that do not wear hardhats and send out alarms to them. As manual inspection is labor-intensive and expensive, it is ideal to handle this issue by a real-time automatic detector. As such, in this paper, we present an end-to-end convolutional neural network to solve the problem of detecting if workers are wearing hardhats. The proposed method focuses on localizing a person’s head and deciding whether they are wearing a hardhat. The MobileNet model is employed as the backbone network, which allows the detector to run in real time. A top-down module is leveraged to enhance the feature-extraction process. Finally, heads with and without hardhats are detected on multi-scale features using a residual-block-based prediction module. Experimental results on a dataset that we have established show that the proposed method could produce an average precision of 87.4%/89.4% at 62 frames per second for detecting people without/with a hardhat worn on the head.

Deep Learning on Construction Sites: A Case Study of Sparse Data Learning Techniques for Rebar Segmentation

“…They conclude that the accuracy of conventional computer vision methods is insufficient to deal with construction site imagery. In contrast, Wang et al [ 1 ] use a CNN to detect hardhats in images using bounding boxes. Their network is trained on a self-collected and self-labeled dataset and achieves high average precision for people with and without hardhats.…”

“…The automation of construction site monitoring is long overdue, especially for progress monitoring, quality inspections and quantity take-offs. These processes rely on visual inputs of either workers or construction site footage and are essential to ensure progression, quality, safety and productivity on site [ 1 , 2 , 3 ]. Currently, the inspections themselves and the analysis of the footage are performed manually, which is labor intensive and requires highly skilled personnel.…”

100+ FPS detector of personal protective equipment for worker safety: A deep learning approach for green edge computing

“…Zhang et al [ 16 ] proposed an improved weighted bi-directional feature pyramid network (BiFPN) to fuse multi-scale semantic features for helmet detection with good results. Wang et al [ 17 ] employed the MobileNet model as the backbone network, proposed a top-down module for enhanced feature extraction, and used a residual-block-based prediction module for the helmet detection for multi-scale features. Filatov et al [ 18 ] designed an automatic helmet monitoring system for surveillance cameras based on MobileNet, which can meet the demand for real-time detection but there is still room for improvement in detection accuracy.…”

Hard hat wearing detection based on head keypoint localization

“…Nath et al (2020) compared three different approaches to detecting PPE based on YOLOv3, including the following: detecting PPE and people to then establish workers -PPE relationship based on bounding box relative position, detecting PPE wearers and non-wearers directly and finally detecting only people to determine if they are wearing PPE with different model. Wang et al (2020) focused on real-time processing with MobilNet (Howard et al, 2017) architecture. Zhou et al (2021) tested YOLOv5 for this application.…”

“…This problem is well known in subcategory classification (Cai et al, 2017;Luo et al, 2019;Han et al, 2019) as it is harder to develop a model that can correctly distinguish fine details between subcategories, in this case the presence of a hard hat. It is for this reason, the best-performing models in this group actually look for the human head instead of the whole person (Fang et al, 2018b;Wu et al, 2019a;Wang et al, 2020;Zhou et al, 2021), making them less suitable for direct transfer learning from a well-trained person detection models.…”