Team Completes The First Data Recording Session for Risky Movement Detection
Team SoterCare successfully wrapped up its first structured data recording session for the "Risky Movement Detection" model, with fellow IITians volunteering to wear our thigh band and walk us through real, everyday movement. The session is a key milestone in building a dataset robust enough to recognize pre-fall behavior before it ever becomes a fall.
We put out a call for volunteers to help us build something we believe elderly care has been missing, a way to act before a fall happens, not just react after one. Most alarm systems only ring once someone has already hit the ground. Our goal with SoterCare has always been prevention and prevention starts with data.
That call became a full recording session. Using SoterCare Studio, our custom built data capture tool, volunteers were fitted with thigh-worn IMU sensors recording at 50Hz while they moved through a guided sequence of everyday actions. Each session was logged, labelled, and reviewed in real time so we could catch noisy or incomplete recordings before they ever reached the dataset.
Volunteers were guided through five core movement states that make up the backbone of our gait model: walking, standing idle, sitting idle, standing up, and sitting down. What stood out most was how willing our fellow students were to give up their time for this. Every recording, every repeated "stand up, sit down, walk a few steps and back" was a small but real contribution to a dataset that will eventually train a model meant to protect someone's grandparent. We're genuinely grateful for that.
This dataset now feeds directly into our Proactive Gait Analysis Model, the on-device classifier that pairs a risky transition prediction with a silent haptic warning. The more varied and well labelled our data, the sharper that model gets at telling a normal wobble apart from a genuinely risky one.
We're now moving into the next phase cleaning, labelling, and feeding this new batch of recordings into training. We'll be sharing results as the model improves. As always, this milestone belongs to everyone who showed up, sensor strapped to their thigh, and walked back and forth across a room for data collection.
With thanks to every volunteer who contributed.