Large, Complex, and Realistic Safety Clothing and Helmet Detection: Dataset and Method

TL;DR

This paper tackles the practical gap in industrial safety monitoring by introducing SFCHD, a large-scale, realistic dataset for safety clothing and helmet detection collected from chemical plants, and presenting SCALE, a plug-in low-light enhancement module that leverages spatial and channel attention. SFCHD provides 12,373 images, 50,552 annotations across 7 categories, and diverse lighting scenarios to stress-test detectors. SCALE improves detection under challenging illumination, achieving measurable gains on both public low-light benchmarks and the SFCHD dataset, and analyses demonstrate the complementary value of SAP and CAP pathways. The work contributes a valuable resource for industrial safety perception and a versatile enhancement tool with potential to improve real-world deployment of object detectors in harsh environments.

Abstract

Detecting safety clothing and helmets is paramount for ensuring the safety of construction workers. However, the development of deep learning models in this domain has been impeded by the scarcity of high-quality datasets. In this study, we construct a large, complex, and realistic safety clothing and helmet detection (SFCHD) dataset. SFCHD is derived from two authentic chemical plants, comprising 12,373 images, 7 categories, and 50,552 annotations. We partition the SFCHD dataset into training and testing sets with a ratio of 4:1 and validate its utility by applying several classic object detection algorithms. Furthermore, drawing inspiration from spatial and channel attention mechanisms, we design a spatial and channel attention-based low-light enhancement (SCALE) module. SCALE is a plug-and-play component with a high degree of flexibility. Extensive evaluations of the SCALE module on both the ExDark and SFCHD datasets have empirically demonstrated its efficacy in enhancing the performance of detectors under low-light conditions. The dataset and code are publicly available at https://github.com/lijfrank-open/SFCHD-SCALE.

Figures (6)

• Figure 1: Examples of images from existing datasets and the SFCHD dataset. Here, (a)-(e), (f)-(j), and (k)-(m) are images selected from the Pictor-v3, SHWD, and SFCHD datasets, respectively.
• Figure 2: The developmental trajectory of object detection methods based on deep learning. The upper part illustrates the one-stage methods and the lower part showcases the two-stage methods.
• Figure 3: Examples of images with different lighting conditions. Here, (a)-(c), (d)-(f), and (g)-(i) are the images based on normal-light, high-light, and low-light, respectively. In (d)-(f), it is extremely difficult to detect whether workers are wearing safety clothing and helmets because the surrounding light is too high. Similarly, in (g)-(i), too low light around workers makes it difficult to conduct object detection tasks.
• Figure 4: The proportion of instances per category in the SFCHD dataset.
• Figure 5: The quantity distribution under different lighting conditions in the SFCHD dataset.