Part II: The Positive Impact of Computer Vision and Machine Learning on Ergonomics
In Part I of this two-part article, we reviewed how computer vision and machine learning can be used in ergonomics. In Part II we focus on the positive impact these technologies can have on the practice and industry.
As discussed in Part I, a corporate ergonomist, ergonomic consultant or risk control consultant can use the visualized outputs to identify, communicate and justify ergonomic improvements that will protect worker safety.
With the typical manual observation approach currently used, it takes roughly 20 minutes to an hour to evaluate a one-minute task. And determining postural angles in common tools such as RULA, REBA and the NIOSH Lifting Equation is one of the hardest, risk factors to measure. There are already software applications that automate these tools to avoid having to traverse tables or perform calculations. But even with this advancement, human observation is still required to determine the postural angles. Alternatively, one step in the generation of the “processed” video described above (using computer vision and machine learning) includes the identification of the body joints frame by frame. This postural angle data can be integrated into automated ergonomic software to calculate scores such as RULA, REBA and the NIOSH Lifting Equation.
All you need is a simple camera on a mobile device (no set up required) to record the video; the worker does not have to stop to put markers or sensors onto the body. This avoids disruption and discomfort.
Multi-camera systems are not only disruptive, they are also quite expensive. They have relatively high accuracy but are often plagued with data problems that have to be addressed before the final result. This is not a criticism of this technology but instead a statement that this type of technology is very hard apply in many industrial situations.
Due to the speed and ease-of-use of the computer vision approach described, a safety professional can assess a larger percentage of the tasks performed in a particular facility. Moreover, more time can be spent on solutions and training rather than on assessments.
As we saw before, the graphs and video frames from this technology can help to make a solid case for action. The visualization brings recommendations to life. Quoting a workers compensation consultant, “It starts the conversation.”
Improved Engagement and Training
Workers are intrigued by the skeleton and this new technology – the mere act of introducing technology to the practice of ergonomics elicits more interest; even better, a worker who can see their own skeleton represented in a video makes this very personal for the worker.
Moreover, if a worker can see the risk of the body parts visualized in graphs, we have an opportunity for education and training. For example, a video of a worker doing a job in two different methods is compared in an intuitive way, thereby clearly demonstrating the advantage of the one over the other.
To extend this even further, risk comparison graphs can be used to train new workers. For example, if a video of a more experienced worker is analyzed and then compared to a video of a new worker doing the same task, the video and comparison graphs can help the new worker adopt safer techniques.
Because after all, any new technology must enable and enhance worker participation in reducing ergonomic injuries. Otherwise, new technology won’t have an impact on worker safety.
The promise of computer vision and machine learning can take us way beyond the ergonomic tools we use today. These ergo tools were developed at a time before smartphones were ubiquitous. These tools are paper and pencil lookup tables and linear polynomial equations; that is what we had at the time. And these tools were designed to avoid needing special equipment and be brief enough to be administered quickly as an initial assessment tool.
Ergonomists and safety professionals know these tools have their challenges. For example, REBA and RULA assess one pose at a time. Whereas, with computer vision, you can assess a continuum of risk throughout a video.
These traditional tools also tend to categorize risk factors into ranges. For example, in REBA a neck angle is rated in ranges of 0 to 10 degrees, 10 to 20 degrees and over 20 degrees. Whereas with computer vision, you can measure body angles that don’t have to be grouped into ranges, thereby giving you a more granular result.
But now smartphones are everywhere. And we have computer systems that can quickly process large, complex data sets (like thousands and thousands of images). And we have started developing sophisticated machine learning algorithms that can analyze a video to generate postural data.
One strategy to enable the positive impact of these technologies on the profession is to apply computer vision and machine learning to tools that are already widely used and from this real-life industrial setting experience, chart a journey for a data-intensive but easy and quick way to assess ergonomic risk. We have not really impacted the world with this technology unless it gets used “in the field.”
As the technology improves, computer vision will become even more accurate and will be able to identify more and more risk factors, maybe even to the point of enhanced biomechanical analysis. The ability of this technology to capture posture, duration and frequency data has arrived already. Exciting research is being conducted right now to capture force just from the video. Imagine that.
Furthermore, consider the impact this technology can have on physical therapy provided to workers to prevent an injury or in the process of evaluation for return to work after an injury.
There are so many possibilities.