According to the United States Bureau of Labor Statistics in 2020 there were over 2.8 million non-fatal workplace injuries and illnesses. This is in addition to the thousands of deaths due to workplace accidents that occur in the US every year. Taken together these statistics represent an enormous amount of potentially avoidable suffering.
Employers have a duty to help keep their workers safe, increasingly this responsibility is being honored with the help of computer vision based AI models. With the help of annotated image and video data, integrated AI safety systems can spot poor safety practices, impairment, and even give warning of potential accidents.
In this blog we will look at three promising AI applications that could make workplaces considerably safer in the near future. Professional annotation services, like Keymakr, can support this technology by providing developers with accurate, affordable datasets.
Automated PPE detection
A vital component of worker safety is wearing the personal protective equipment in the right way. Not having the right protective gear, or wearing it incorrectly, can lead to accidents across a range of industries. It can be difficult for supervisors to monitor PPE adherence across large worksites, and this leads to many training opportunities being missed.
AI models can be trained to assess CCTV or still images and recognize when PPE is missing or wrongly deployed. This data can then be used to quantify overall levels of adherence in a particular worksite, allowing managers to intervene before accidents occur. Similar models can also be used to check levels of mask wearing and social distancing in spaces that need to follow Covid-19 protocols.
For these models to be reliable developers need access to annotate image and video. Video annotation can be a particular challenge to accomplish due to the time it takes to correctly label thousands of frames of video. Keymakr has developed annotation tools specifically with video annotation in mind. This kind of specialization can make video annotation a much more straightforward process.
Impairment, due to alcohol, fatigue, sleep deprivation, or drug use, can lead to serious accidents. This usually occurs when impaired individuals operate machinery, on work sites, in factories, and on the road. AI models are being rolled out that have the capacity to identify impairment by analysing the eyes of users.
The thinking behind this approach is that machine operators can use AI powered devices to check if they are capable of safely working. This will prevent impaired workers from taking the wheel and potentially endangering themselves and others.
Detecting impairment by analysing the human eyes requires AI models that are trained with thousands of similar images. Accuracy is essential when safety is the goal, this means that verification processes have to be of a high standard. Keymakr employs three levels of human verification, as well as an additional automated sanity check.
The ultimate goal for workplace safety AIs is an integrated system that can give real-time warnings about potential accidents. This means having full awareness of the location of workers, if they are entering unsafe areas, if they are close to dangerous machinery, and if they are complying with health and safety standards.
This data can provide managers with predictive analytics and help them locate the places where potential accidents could occur. Semantic segmentation annotation can provide these models with the additional information that they need. By dividing training images in pixel classes, it is possible to identify safe zones, and dangerous machinery, giving context to the positioning of workers in images.
Keymakr’s experienced annotation teams and bespoke annotation solutions support computer vision innovation.