Introduction

Artificial Intelligence and computer vision have made a very disruptive impact on the energy, oil, and gas industry over the last decade. The specific use-case of several computer vision applications in industrial safety has seen a tremendous impact, with several institutions rapidly adopting the use of such safety and health monitoring applications rather than the traditional inefficient processes.

In this article, we will quickly understudy some troubling challenges in occupational health and safety, we also take a brief look at the general computer vision landscape, and finally how this technology can be applied to occupational health and safety more specifically.

The occupational health and safety challenges

Very frequent operational downtime and losses in total man-hours due to health and safety-related incidents could be a result of harmful exposure to hazardous substances or frequent violations of safety standards in a facility or an organization, amongst other numerous factors.

The International Labour Organisation (ILO) correctly captures the enormous burden of poor working conditions and HSE (Health Safety and Environmental) related challenges in a study as it identifies some major pain points.

According to the ILO estimates, exposure to Hazardous substances alone accounts for diseases related to death among workers, and it’s estimated to cause around 651,279 deaths a year. Also, the construction industry has a disproportionately high rate of recorded accidents which could be due to in-adherence or disregard for safety and protective equipment.

This study only buttresses the need for a robust framework for accurate monitoring of safety and health operations and regulations in many organizations.

What is Computer Vision & Artificial Intelligence?

Artificial intelligence is no longer considered simply just a buzzword as its widespread impact has now been felt across multiple industries including agriculture, healthcare, and media advertising as notable mentions.

Computer vision on one hand can be described as giving machines the capability to find patterns and extract information following a similar principle as a human vision all from images and videos. Early adopters of computer vision technology have realized a massive impact in factory robotics and autonomous flight and drone technology.

Computer vision adoption in health and safety impact and challenges.

Monitoring compliance to safety standards in terms of gear, outfit, and compliance with facility safety regulations can be challenging and expensive when done manually by humans. Some major pain points across the energy industry may include;

• Unexpected Downtime from failures and Accidents.
• Gas Exposure, Flammable, and leakage.
• Access Violation Monitoring.
• Health Monitoring, Material conditions.

Machine learning computer vision algorithms aim to address these issues through the creation of a healthy workplace facility culture that enforces compliance with HSE protocols, safety violations, personal protective equipment (PPE) wear detection, etc.

A typical operating facility often requires constant surveillance of personnel on sight to ascertain that personnel on duty uphold safety regulations during operations, one common challenge we point out is that human facility regulators might tend to be complacent and inefficient even when trained professionally.

Security cameras and devices on one hand are mechanical in their operations, hence can be deployed efficiently for health and safety facility surveillance.

Generally, 4 common real-world practical applications of computer vision for facility health and safety may include;

• Object detection for personnel safety gear and PPE.
• Personnel counting for capacity restrictions.
• Human activity detection for safety compliance.
• Access region controls.

Applying computer vision systems to most general use-cases including health and safety typical requires a simple workflow that traditionally builds on the following core components;

• Image classification e. g, hard-hat or safety vest image classification.
• Object detection and localization e. g, the specific positions of the several hard-hat images on a video frame.
• Safety logging and alert systems.

Image classification

Image classification is a specific type of computer vision application where different classes or categories of images are classified according to their specific sub-groups example may include sub-groups of hard-hat and safety boats, road cones, gloves, welding masks etc.

This class of computer vision applications can be very fundamental as it tends to solve the single biggest problem of distinguishing one class of item from another. More specifically the problem of image classification works by using a machine learning algorithm to find and generalize hidden patterns in different classes of images when given massive amounts of training data including a training input and output pair.

Object detection

More applicable is the problem of object detection in computer vision which builds on the idea of image classification already discussed. In this application, a real-time video feed may be processed and analyzed to detect different object categories. A common use case of this application may involve detecting and bounding the region of a factory worker on site who is not putting on a hard hat. Such applications and similar use cases that follow similar ideas are the central problem of object detection.

Safety monitoring, alerting, and performance metrics.

Ultimately, the computer vision system developed is integrated into most existing processes and workflows for safety monitoring and alert feedback. This is the most important step in building a computer vision application. In essence, the process of deploying a computer vision application for real-time safety monitoring and the alert system helps to identify practical scenarios for which the application developed either perform well or worst.

For most computer vision applications especially in the case of facility safety and health, depending on the intended use case, business stack holders are entirely at liberty to decide on objective performance metrics for which the effectiveness of the computer vision application can be measured. For facility safety and health, some useful metrics may include;

• Personnel compliance to effective use of safety wear, protective equipment, and clothing.
• A turn-over rate for which total man-hour lost time due to injuries can be reduced.

The above-mentioned metric might not be most suitable for any indented application deployment use case, but hopefully gives a general sense of how to formulate a typical business metric for a computer vision system in safety and health. It is essential to evaluate closely how a particular computer vision application developed can affect the business key performance index (KPI) in a typical business outfit.

Conclusion

In the current post, we have examined closely some troubling statistics around occupational health and safety to further buttress why the topic of safety is such an important discussion. Very important in this current post we examined the effectiveness of adopting more artificial intelligence and machine learning technology use cases for computer vision in health and safety.

We also briefly discussed several aspects of a computer vision system from conceptualization to production and monitoring. Formulating a good deployment metric for when to know if a computer vision application is adding the right business value was also discussed. With more adoption of artificial intelligence computer vision applications, we can expect to reduce downtime during facility operations and fewer unsafe conditions that may lead to loss of life and property.