This blog post was originally published at NVIDIA’s website. It is reprinted here with the permission of NVIDIA.
By 2030, more than one in five Americans will be 65 or older, becoming the United States’ largest group of seniors ever.
Silicon Valley-based startup Butlr has developed an AI platform designed to keep seniors safe while preserving their privacy.
Their AI-based platform uses a neural network to interpret different temperature data that its sensors, which are strategically placed in elder care facilities, pick up.
The model then creates real-time infrared images of people that, while largely featureless and blurry, are sufficiently detailed to help aides at eldercare facilities keep close tabs on patients.
“What we’re trying to do is leverage temperature data at the edge to save lives, using intelligence that keeps personal information private,” said Honghao Deng, Butlr’s CEO and co-founder.
Butlr’s designed its AI solutions to collaborate with, rather than replace, elder care aides.
The model can detect if a patient has fallen down, or remained in bed for too long, and automatically alerts caregivers if it identifies an emergency. Additionally, if a patient repeatedly gets out of bed to use the restroom the AI can flag this behavior for caregivers, offering potentially early detection of health issues, like urinary tract infections, that might otherwise go undiagnosed for days.
At the end of each day, the model summarizes patients’ activities, translating heat patterns into behavioral summaries—offering heat signature-to-text capabilities. These detailed entries can free up caregivers to spend the end of their shifts engaging with patients rather than writing up patient summaries.
Butlr’s biggest innovation is repurposing infrared heat signatures, which, because they’re relatively low resolution, can be continuously captured by inexpensive, battery-powered sensors.
These lightweight sensors are easily installed in virtually any room. Their batteries last around seven years, which means the overall system is flexible and inexpensive to install while providing comprehensive temperature maps of an elder care facility.
Critically, the sensors integrate with a robust model running in the background, which powers real-time inferencing interpreting patients’ behavior.
To analyze the over one billion frames per day and around two petabytes of data per month, Butlr uses NVIDIA GeForce RTX 2070 cards for data visualizations. In the AWS cloud, it runs Amazon SageMaker ml.p5 instances powered by NVIDIA H100 Tensor Core GPUs for training and ml.g5 instances with NVIDIA A10G Tensor Core GPUs for real-time inference.
The model segments interior spaces into two-foot squares and takes around 10 temperature readings per second of each square. The model then reconstructs that low-resolution heat data into coherent, real-time temperature-defined images, which gives caregivers enough information to monitor seniors’ behavior while not revealing—or recording—personal information.
Video 1. Lightweight sensors turn heat signatures into real-time temperature maps to safely monitor activity
Another key advantage of temperature signatures is what they mean for privacy.
Temperature data is far less visually precise than video. So, while data relevant to keeping seniors safe is captured, their likenesses are not.
“Our models process high frame rates and low resolutions. There’s no way to even see a person’s limbs, let alone a face,” Deng said. “With our model, we’re able to reconstruct important shapes and understand what someone is doing, without revealing someone’s identity.”
Going forward, Deng sees the need for elder care solutions only growing.
“All of us just refuse to admit we’re getting old,” he said. “But we’re all getting old, and we’re going to need to create innovative solutions because unless things change, there’s just not enough trained caregivers to help everyone.”
Read more about Butlr’s temperature mapping technology and its potential widespread use cases.
Related resources
- GTC session: Enhance Patient Safety With Privacy-Preserving AI: Multimodal DNN and VLM for Event Detection in Healthcare
- GTC session: Secure Your Data and Model IP While Improving Productivity
- GTC session: Protect and Control Your AI Applications (Presented by Cloudflare)
- NGC Containers: ASR Parakeet CTC Riva 1.1b
- NGC Containers: Earth-2 CorrDiff
- NGC Containers: NMT Megatron Riva 1b
Elias Wolfberg
Executive Communications team, NVIDIA
