AI-Driven Detection of Personal Protective Equipment Usage in Occupational Settings: Enhancing Workplace Safety

YOLO-NAS based Intelligent Workplace Safety Monitoring System for Real-Time Personal Protective Equipment Compliance Detection

Workplaces in construction and industry suffer from a significant number of workplace accidents because of a lack of safety mechanisms like helmets and high-visibility vests. In order to solve this issue, an automatic detection system for detecting the presence of the worker's helmet & safety vest using a deep learning model is developed. You Only Look Once – Neural Architecture Search (YOLO-NAS) algorithm was chosen to be used in the model because of its fast and high-quality detection process. A dataset with images containing healthy workers wearing helmets & safety vests (with some workers appearing in different safety vests) is collected and used as the training set. The resulting model is then saved for future usage in making predictions with new images. Then, this model is connected to Streamlit, which provides a convenient Web-based user interface through which a user can provide an image as an input to the model. The trained neural network is applied to the input image, and all detected objects that correspond to helmet & safety vest are put in bounding boxes on the input image. Moreover, each of the bounding boxes is labeled with the name of the object and confidence score of the detection. Thus, a clear visualization of the safety equipment worn by the worker is provided. Overall, the developed detection system significantly decreases the human intervention needed for visual verification of the safety helmet or vest. The described system is a user-friendly and cost-effective way of evaluating workplace safety compliance.

Sensor-Based and AI-Driven Ergonomic Seated Posture Detection for Workplace Risk Prevention

Background: Work-related musculoskeletal disorders (WMSDs) remain one of the most prevalent occupational health problems worldwide. To prevent the development of these WMSDs in an office space, a chair designed for office monitoring capable of accurately identifying ten representative seated postures and measuring environmental factors was developed and validated. Methods: To evaluate office working conditions, the chair has three embedded Printed Circuit Boards (PCBs): one directed towards seat pressure management, one directed towards environmental measurements and one PCB to manage the entire system. Machine learning approaches were then applied to establish a model that effectively predicts the seated position. The environmental data were also analyzed. Results: The seated position classification presented an accuracy of 80.99% in controlled conditions, while in a real-world context the accuracy was 65.98%. The environmental management showed low errors, except for the PM2.5 and PM10, with relative errors above 30%. Conclusions: This work presents an initial promising first step for an ergonomic office management solution.

Lightweight Six-Class Personal Protective Equipment Detection in Construction Scenes Using YOLO11n

Personal protective equipment (PPE) detection is a critical task in construction safety management, yet real-world construction scenes are often affected by occlusion, scale variation, complex backgrounds, and small-object interference. To evaluate the applicability of lightweight object detection for practical site monitoring, this study investigates a six-class PPE detection task in construction scenes based on YOLO11n, covering person, hard hat, vest, safety boots, gloves, and mask. Experiments were conducted on a public PPE dataset containing 1,617 images, which were divided into 1,100 training images, 304 validation images, and 213 test images. The model was trained with an input size of 640 × 640 for 50 epochs and evaluated using Precision, Recall, mAP@0.5, mAP@0.5:0.95, and inference speed. The results show that YOLO11n achieved Precision, Recall, mAP@0.5, and mAP@0.5:0.95 values of 0.821, 0.872, 0.846, and 0.566, respectively, on the test set, while maintaining an inference speed of approximately 103 FPS on an RTX 4050 Laptop GPU. Class-wise analysis indicates that person, hard hat, and vest were detected more reliably, whereas mask, gloves, and safety boots remained more challenging because of their small scale, occlusion sensitivity, and ambiguous visual features. Overall, the results demonstrate that lightweight YOLO11n provides a reasonable balance between detection accuracy and real-time performance for six-class PPE monitoring in construction environments, and it can serve as an effective baseline for future improvement in small-object enhancement and deployment-oriented optimization.

Deep Learning-Based Worker Posture Classification for Ergonomic Risk Evaluation in Manufacturing

Continuous ergonomic monitoring in manufacturing remains challenging because conventional posture assessment methods rely on manual observation, making evaluations time-consuming, subjective, and unsuitable for continuous industrial applications. This study proposes an automated ergonomic risk assessment framework that integrates Media Pipe Pose with a Convolutional Neural Network (CNN) to classify worker postures into low-, medium-, and high-risk ergonomic categories. The framework extracts 33 anatomical body landmarks from RGB images and video frames to generate marker less posture representations for deep learning-based classification. A dataset consisting of 4,500 posture samples collected from assembly, packaging, and welding workstations was expanded to 12,000 samples through data augmentation techniques, including rotation, scaling, horizontal flipping, and brightness adjustment, to improve model robustness and generalization. The CNN model was trained and evaluated using an independent test dataset, achieving an overall classification accuracy of 94.2%, with precision, recall, and F1-score consistently exceeding 94% across all ergonomic risk categories. Comparative evaluation against a conventional REBA/RULA-based rule-driven assessment demonstrated that the proposed framework improved classification accuracy by 7.5 percentage points while eliminating the need for manual posture scoring and reducing observer subjectivity. Furthermore, computational performance analysis showed that the complete inference pipeline operated at an average of 14 ms per frame (approximately 28.5 FPS) on a standard Intel Core i7 CPU with 16 GB RAM, without requiring GPU acceleration, indicating its suitability for real-time deployment in manufacturing environments. The proposed Media Pipe–CNN framework provides an efficient, accurate, and marker less solution for automated ergonomic risk assessment, supporting intelligent occupational safety management, continuous workplace monitoring, and the implementation of smart manufacturing systems aligned with Industry 4.0 initiatives

Personal protective equipment detection method based on knowledge distillation

A Data-Driven Real-Time Fall-from-Height Detection Method for On-Device Worker Safety Wearables

Fall-from-height (FFH) detection is a critical component in wearable safety systems, particularly in environments where high-intensity movements can lead to frequent false positives. Conventional approaches based on simple thresholding of acceleration signals often fail to reliably distinguish FFH events from non-fall activities due to overlapping signal characteristics. This paper proposes a data-driven FFH detection method that integrates multiple complementary features into a unified score-based model. The proposed approach first performs structured peak detection to extract candidate impact events while significantly reducing the number of samples requiring further processing. Each candidate is then evaluated using pre-peak structure, post-impact stability, and pressure variation, which respectively capture structural, temporal, and physical characteristics of FFH events. Based on statistical analysis, feature-wise score contributions are designed to reflect their discriminative strength, and the final FFH decision is performed using an additive scoring mechanism. This formulation enables flexible handling of ambiguous cases while preserving strong FFH characteristics. Experimental results demonstrate that the proposed method maintains 100% recall at the selected decision threshold while significantly reducing false positives from non-FFH activities. In addition, the peak detection stage reduces more than 99% of raw samples, enabling efficient on-device processing suitable for wearable systems. The proposed method also includes quantitative analysis of latency characteristics. Although FFH inference latency is influenced by asynchronous pressure sensing, the delay remains bounded and predictable, and most detections are completed within a practical time range for real-time wearable safety applications. Overall, the proposed method achieves a practical balance between detection sensitivity, false-positive suppression, computational efficiency, and real-time feasibility, demonstrating its applicability to wearable safety systems.