NIT Rourkela develops multi-sensor AI system to track sleep posture

Mumbai: Poor sleeping posture can cause long-term health conditions by putting sustained, uneven pressure on the spine, joints, and nerves for hours at a time. Even for a physically active person, this can lead to chronic musculoskeletal pain, spinal degeneration, obstructive sleep apnea, nerve damage, poor digestion and acid reflux, and arthritis, among others. For bedridden patients, poor posture can cause serious complications such as pressure ulcers or bedsores.

However, now comes an AI-enabled system that can track human sleep postures - thanks to the efforts of National Institute of Technology-Rourkela.

This is useful in healthcare settings, as it provides a non-intrusive way to monitor patients while maintaining their privacy, even when they are covered with a blanket.

The findings of the NIT-Rourkela research have been published in the IEEE Sensors Journal. 

The paper is co-authored by Prof Saptarshi Chatterjee, Assistant Professor, Department of Electronics and Communication Engineering, along with his B.Tech student  Shiladitya Mondal at NIT-Rourkela and Dr. Debangshu Dey, Assistant Professor, Department of Electrical Engineering, Jadavpur University.

According to the press note released by NIT Rourkela, currently, patient posture monitoring is mostly done manually, which can be inconsistent and prone to human error. Although wearable sensors can be an option, but they are uncomfortable and costly for the end users. Additionally, some systems use cameras, but they can be limited by factors including insufficient lighting, obstructions such as bed covers, and privacy issues.

“Our system uses generative AI and a fusion approach using low-wave infrared, depth, and pressure map data to identify sleeping positions without the direct use of RGB images. The model works well in a variety of difficult situations including low-light situations and different types of cover,” said Prof Chatterjee.

By combining heat-based imaging, body shape data, and pressure information, the system can deliver accurate results. Lab experiments show that this no-contact model achieves around 98 per cent accuracy, making it reliable for real-world use.

The automated system can reduce the workload of caregivers and allow continuous monitoring. At the same time, since it does not rely on visual imaging, it helps protect patient privacy.

Prof Chatterjee and his team developed an AI-based system that uses three types of sensors: an imaging sensor in the long-wave infrared spectrum that captures and uses body heat to track sleep postures, and does so without using visual data, even if the subject is covered by a blanket, a depth sensor that captures the shape and posture of a person and a pressure sensor that assesses how the weight is distributed on the bed.

To process the data from these sensors, the team developed a generative AI model to represent the human body, and a graph-based neural network to classify the various postures of body joints.

To be used as an integrated module with multi-modal imaging systems, the approximate cost of the developed technology will be Rs 30,000, with scope for further reduction with mass-scale development.

As a next step, the research team plans to extend the usage of this technology for identification of specific incorrect sleep posture-related health issues and other diseases.

With further refinement and real-world testing, the technology can move closer to practical deployment across healthcare settings. Its scalability and adaptability will also open possibilities for applications beyond hospitals, including home-based care.