In a breakthrough that could transform patient care, researchers at the National Institute of Technology (NIT), Rourkela, have developed an AI-powered system capable of accurately tracking sleep postures, even under blankets, without compromising privacy.
The research findings, published in the IEEE Sensors Journal, are expected to significantly improve patient monitoring in hospitals and home-care settings. The innovative AI system appears to have addressed a long-standing challenge in healthcare — monitoring patients continuously without intrusive cameras or uncomfortable wearable devices, while ensuring high accuracy and dignity.
The NIT Rourkela team has designed a non-intrusive system that combines three complementary sensing technologies — long-wave infrared (LWIR) sensors, depth sensors and pressure sensors. The LWIR sensors can capture body heat without revealing visual identity, while the depth sensors map body shape and posture. The pressure sensors track weight distribution on the bed.
“These inputs are processed through an advanced AI pipeline that integrates a generative model and a graph-based neural network. Our system leverages generative AI and multimodal sensor fusion to detect sleeping postures without directly using visual images. It first fuses multimodal data, then reconstructs a clear representation of the human body before identifying joint positions and overall posture,” Prof Saptarshi Chatterjee, co-author of the research, said.
Poor sleeping posture is often recognised as a major contributor to long-term health complications. Several studies indicate that sustained uneven pressure on the spine, joints and nerves during sleep can lead to chronic musculoskeletal pain, spinal degeneration and conditions such as obstructive sleep apnoea. Incorrect posture can also cause pressure ulcers, commonly known as bedsores, for bedridden patients.
Currently, patient posture monitoring is mostly done manually, which can be inconsistent and prone to human error. Wearable sensors are another option, but they are often expensive and uncomfortable for patients. Existing camera-based systems also face challenges such as low lighting, obstruction due to blankets and privacy concerns, making them less suitable.
According to the new research, the AI model can achieve an accuracy of 98.46 per cent, outperforming existing state-of-the-art systems in in-bed pose estimation. By combining heat-based imaging, body shape data, and pressure information, the system can deliver accurate results.
The system has the ability to function effectively under real-world conditions, including low lighting and heavy occlusion caused by blankets. Unlike conventional camera-based monitoring, the system does not rely on RGB imaging and ensures patient privacy.
“The automated nature of the 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,” said Debangshu Dey, another co-author.
The technology has been designed for seamless integration into hospital beds and home-care systems, enabling continuous, automated monitoring of patients, elderly individuals, and those suffering from sleep disorders such as sleep apnoea.
The estimated cost of the system is around ₹30,000, and it could be reduced further through large-scale production. Its automated nature could significantly reduce the workload on caregivers while improving the accuracy of patient observation.
Researchers have planned to expand the system’s capabilities to identify specific posture-related health risks and detect disease conditions linked to sleep behaviour. The study also highlights the potential of such multimodal AI systems in broader healthcare applications, including fall-risk assessment and seizure monitoring.
“With further refinement and real-world testing, the technology can move closer to practical deployment across healthcare settings. While the current model requires substantial computational resources, future work will focus on optimisation techniques to enable deployment in resource-constrained environments,” said Shiladitya Mondal, a BTech student.
Sleep posture monitoring through AI system
