The Korea Institute of Industrial Technology (KITECH) said on the 30th that it developed an artificial intelligence (AI) technology that precisely predicts muscle load using only camera footage.
In industrial settings, tasks that involve repetitive motions and heavy loads increase workers' muscle fatigue and the risk of musculoskeletal disorders. To address this, electromyography (EMG) sensors, which accurately measure muscle load, are used, but they are uncomfortable to wear for long periods and are affected by sweat and work clothes, making them difficult to use on site.
A team led by Senior Researcher Tae Hyun-cheol at KITECH's Manufacturing AI Research Center developed a solution that analyzes work footage to calculate the current usage ratio relative to maximum muscle strength. Through repetitive task experiments, the team built a dataset of video and electromyography signals and trained AI to analyze differences in muscle load according to human physical characteristics and work conditions.
The team also advanced the model with technology that can analyze how work motions change over time. By combining a recurrent neural network that recognizes temporal flow with a multilayer neural network structure, the model can predict changes in muscle load throughout the entire sequence of tasks that continue consecutively, such as lifting, moving, and setting down.
To verify the performance of the developed model, the team conducted experiments that replicated a work environment. They attached electromyography sensors to participants to measure actual muscle load and compared and analyzed those measurements with the AI model's predictions, finding values that were nearly identical to the sensor readings.
Senior Researcher Tae Hyun-cheol said, "Video-based muscle load estimation is an approach that goes beyond the discomfort of wearing sensors," adding, "We will expand it into a general-purpose model that can manage workers' fatigue and load under various work conditions in manufacturing, logistics, and healthcare."