Overview
I worked with three teammates on this knee brace for my NYIT senior design project.
We explored whether detecting a fall and limiting knee bending could help reduce
leg injuries in older adults.
We chose a preset bend limit because suddenly locking the leg raised stability
concerns. The goal was to limit overbending while leaving room for the user to
adjust their leg and regain balance with a mobility aid.
The prototype uses a Random Forest model to detect falls and two push-pull
solenoids to engage the lock. It releases after five consecutive seconds of
non-fall readings. The design relies on a mobility aid to help the user regain balance.
This is an early prototype for future students to build on. A useful next question
is how the detection and locking behavior hold up across different users and movements.
Pictures
Brace viewed from the side
Sensor on the chest
Sensor on the hip
Demo
Fall Detection Results
In tests using my own movements, the fall detector reached 94.5% accuracy with
a 300 ms sampling interval. That interval gave the best result among those I tested.
The model used 1,000 training samples for each label: falling and non-falling.
The configuration used 30 decision trees per label, each with a depth of 10.
The two experiments and their confusion matrices are shown below.
Confusion Matrix
Code and Setup
The code and setup instructions are in my GitHub repository.