Overview / Usage

Construction is one of the most dangerous industries, and ensuring safety at construction sites is of utmost importance. Hard hats and safety vests are critical components of worker safety gear, providing a crucial first layer of protection against severe injuries. Construction workers are exposed to many hazards, including falls and head injuries. We propose a safety gear monitoring project for construction workers to tackle these concerns. The project involves the development of a computer vision system that can detect and track safety gear worn by construction workers, such as helmets and vests. The system will use deep learning techniques to identify and classify safety gear and track its usage in real-time. Our project's novelty lies in the system's ability to alert workers when they are not wearing the necessary safety gear or if their gear is damaged or expired. Moreover, we will design a user-friendly interface that will be easily accessible to construction workers, supervisors, and safety officers. To develop the system, we will collect a dataset of images and videos of construction workers wearing safety gear such as helmets and vests, and then utilize the powerful tools available in the Intel one API platform. These images and videos will be used to train the deep learning algorithms. The proposed project can significantly enhance the safety and protection of construction workers by guaranteeing that they wear proper safety gear. This will reduce the risk of accidents and injuries, especially those related to falls and head injuries.

Methodology / Approach

METHODOLOGY:

1. Dataset Collection: We collected a dataset of images and videos of construction workers wearing safety gear such as helmets and vests. The dataset include a wide range of conditions, such as different lighting, angles, and environments. The dataset has been annotated, and bounding boxes were labeled around safety gear components.
2. Data Preprocessing: We preprocessed the data to ensure that the images are of high quality and ready for training. We also divided the dataset into training, validation, and testing sets to evaluate the model's performance.
3. Deep Learning Model Development: We've used deep learning techniques to develop a computer vision system that can detect and track safety gear worn by construction workers. We've used the powerful tools available in the Intel one API platform to create a convolutional neural network (CNN) model for object detection and classification.
4. Model Training: We trained the deep learning model using the annotated dataset.
5. Evaluation: We evaluated the system's performance by measuring its accuracy using the testing dataset.
6. Deployment: We deployed the safety gear detection system to ensure workers' safety and compliance with safety regulations.

STANDARD TECHNIQUES:

Computer Vision:

Computer vision techniques will enable the system to detect safety gear worn by workers in real-time, ensuring compliance with safety regulations and reducing the risk of accidents and injuries. The system's deep learning algorithms will be trained on a dataset of images and videos of construction workers wearing safety gear to achieve high accuracy.

Intel oneAPI platform:

Toolkit used: Intel® AI Analytics Toolkit (AI Kit) - Python 3 (Intel® oneAPI 2023.0)

   We have successfully utilized the Intel® AI Analytics Toolkit to optimize our model and achieve superior results. This comprehensive toolkit enables us to accelerate end-to-end data science and machine learning pipelines using Python* tools and frameworks. Leveraging state-of-the-art deep learning frameworks such as PyTorch and TensorFlow, which are optimized for the Intel architecture by the oneAPI platform, has allowed us to achieve high performance and accuracy in our Safety Gear Detection System for Construction Workers. Additionally, the Intel® Extension for Scikit-Learn has been enabled for improved performance. The toolkit also provides support for several pre-trained models, including DenseNet, ResNet, YOLOv3, and more, which have been instrumental in our project. Furthermore, leveraging transfer learning with pre-trained models on Intel DevCloud for oneAPI has boosted our system's accuracy and performance. The detection part of the safety gear is performed on Intel DevCloud, showcasing the versatility and capabilities of the Intel® AI Analytics Toolkit in delivering optimal performance.

MODEL:

DenseNet169

DenseNet is a convolutional neural network (CNN) that connects each layer to every other layer in a feed-forward fashion. This architecture allows for better feature reuse and efficient memory usage.

ResNet50

ResNet, on the other hand, is a CNN architecture that uses residual connections to address the vanishing gradient problem that can occur in deep neural networks. These residual connections allow information to flow directly through the network, bypassing certain layers and making it easier to train very deep neural networks.

FRAMEWORK:

Streamlit

Streamlit is a popular Python framework for building and deploying interactive data science web applications. In our safety gear detection system project, Streamlit can be used for the deployment of the computer vision application.

Streamlit's easy-to-use interface allows for the creation of an intuitive dashboard that displays the results of the computer vision analysis, including the detection of safety gear in real-time. Streamlit's capabilities in handling media files make it an ideal choice for our project, as it can easily handle image and video uploads and even webcam enabling option for real-time monitoring.

By deploying our safety gear detection system using Streamlit, you can make it accessible to a wider audience, including construction workers, supervisors, and safety officers, and provide an efficient way to ensure compliance with safety regulations on construction sites.

Technologies Used

• Intel oneAPI
• Computer Vision
• DenseNet169
• ResNet50
• Streamlit
• TensorFlow
• SkLearn
• Pandas
• NumPy
• Keras

Repository

https://github.com/jeyasri-senthil/SafetyGearDetection_IntelOneAPI

Collaborators