Real-Time Personal Protective Equipment Non-Compliance Recognition on AI Edge Cameras

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

 

 

1.      What are the primary motivations stated for developing an automated safety monitoring system in construction sites. Please explain this in the introduction

2.      How does the research propose to address the current   safety monitoring systems.

3.      It is suggested to include this article in the introduction

Elhanashi, A.; Dini, P.; Saponara, S.; Zheng, Q. Integration of Deep Learning into the IoT: A Survey of Techniques and Challenges for Real-World Applications. Electronics 2023, 12, 4925. https://doi.org/10.3390/electronics12244925

 

 

4.      How does the proposed system utilize CCTV cameras for safety surveillance, Please add paragraph for this explination  

 

5.      What were the specific classes for the dataset used for training the proposed model  

 

 

6.      Explain more the  pre-processing techniques on the  the images  

7.      Why was SSD MobileNet V2 chosen as the backbone architecture for the object detection model. You have to explain more why this model is chosen in comparison to the state of art

8.      What challenges were encountered during the deployment of the model on edge devices?

 

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Reviewer 2 Report

Comments and Suggestions for Authors

The paper "Realtime Personal Protective Equipment Non-Compliance Recognition on AI Edge Cameras" by Glenn Neuber et al. addresses a significant issue in the construction industry: the frequent non-compliance with personal protective equipment (PPE) usage, such as hard hats and safety vests. The authors propose a deep learning-based pipeline for real-time detection of PPE non-compliance, leveraging AI-enabled edge cameras to enhance workplace safety. This research offers a promising solution for improving safety compliance in the construction industry through the use of AI-enabled edge cameras. While there are areas for improvement, the proposed model demonstrates significant potential in enhancing real-time PPE detection and overall workplace safety. While the study makes several significant contributions, there are a few areas where it could be improved and expanded upon. Below are some critical points and suggestions for further research:

1. Dataset Limitations: The dataset primarily comprises images downloaded from the internet, leading to a potential mismatch with real-world conditions on construction sites. The significant imbalance in the dataset, especially the over-representation of images from China, could affect the model’s generalizability and fairness across different populations and environments. Limited diversity in the dataset, particularly concerning different lighting conditions, weather variations, and types of PPE, may hinder the model's robustness and performance in varied real-world scenarios.

2. Model Performance in Adverse Conditions: The model’s performance drops in low-light or humid conditions, which are common on construction sites, especially during early morning or late evening hours, and in certain climates. High resolution images increase processing time, affecting real-time performance, which is crucial for immediate hazard detection and mitigation.

3. Similar Headgear Detection: The model struggles to distinguish between hard hats and other types of headgear, such as hoodies, caps, or beanies. This limitation can lead to false positives or negatives, reducing the system's reliability.

4. Bias and Fairness: The lack of representation in skin tones and other demographic factors raises concerns about the model's fairness and bias, potentially impacting its applicability in diverse workforces. Measures to ensure the model does not perpetuate existing biases are briefly mentioned but not elaborated upon, leaving room for improvement in this critical area.

Suggestions for Further Research are mentioned below: 

1. Dataset Enhancement: Collect and include a larger and more diverse set of images directly from real-world construction sites, capturing various lighting conditions, weather scenarios, and different types of PPE. Increase the representation of diverse demographic groups to ensure fairness and reduce potential biases in the model.

2. Improving Model Robustness: Develop and integrate techniques to enhance the model’s performance in low-light and adverse weather conditions, such as using infrared or multi-spectrum images. Implement image resolution optimization methods to balance processing speed and accuracy, ensuring real-time performance is maintained without compromising detection quality.

3. Advanced Headgear Classification: Enhance the model’s ability to differentiate between hard hats and other similar headgear by incorporating additional features or employing more sophisticated classification techniques. Include training data with various types of headgear and different stages of wearing and removing the headgear to improve the model’s accuracy and robustness.

4. Bias Mitigation and Fairness: Conduct a detailed study on the model’s performance across different demographic groups and environments to identify and address any biases. Implement and test various fairness-enhancing techniques, such as re-sampling, re-weighting, or adversarial debiasing, to ensure the model performs equitably across diverse populations.

5. Integration and Real-World Testing: Perform extensive real-world testing and validation of the model on actual construction sites to evaluate its practical applicability, reliability, and impact on safety outcomes. Explore integration with other safety systems and sensors, such as fall detection, to create a comprehensive safety monitoring solution.

 While the proposed model for real-time PPE non-compliance detection using AI-enabled edge cameras shows great promise, addressing the identified limitations and areas for improvement through further research and development will be crucial. Enhancing the dataset, improving model robustness, addressing bias and fairness issues, and conducting extensive real-world testing will ensure the solution is reliable, equitable, and effective in diverse construction site environments.

Author Response

Please see the attachment.

Author Response File: Author Response.doc

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Thanks author