Discount-Cyborg

Bootstrapping deep learning for fall risk evaluation

Comment

Inspiration

People fall, we don't want them to. The elderly are often at a higher risk of falling, we thought that the current diagnosis methods that often relied heavily on the doctor's own vision and judgment could be improved, as the risk of misdiagnosis is very high.

Furthermore, most of us can only get a health care professional a handful of times. If the way we walk or carry ourselves changes in between visits we could be at a risk of falling while having a clean bill of health. With this code all the sensors needed can be found in a smartphone and diagnoses is only an app away.

What it does

The device allows for the bootstrapping of the data generation process to allow for development of deep learning algorithms. The non-consumer hardware involved simply allows for the augmentation of the accelerometer data by providing a grounded truth output for the data.

How we built it

We wired 2 buttons onto each shoe to use as pressure sensors, connected them to an arduino. The data from the buttons was processed in python to obtain the different parts of a walk: swing, stance, and stride. We made an android app to collect accelerometer data from a phone and add it to Firebase.

Challenges we ran into

In early version of this project we decided that a pressure sensor would be needed for this hack. Void of any such sensor we endeavored to create one. We noticed that a change in capacitance can be measured when a change of distance between the conductive plates occurs. To map pressure into a linear change in thickness of the capacitor we created a capacitor using puck shaped stress-ball pieces as a dielectric and cut up pieces of pop can. This proved very hard to fabricate, and then calibrate and we ended up making an executive decision to proceed with a different approach.

Mounting the sensors to the human body also proved quite difficult and finding a way to lay the buttons down so that they are level with the user's shoe was also hard.

What we learned

Hardware is hard……things will break at the most inconvenient times and nothing is as easy as you thought it would be. It turns out you cannot build a capacitor with a stress ball and a pop can.

What's next for Discount-Cyborg

Train the deep learning models to extract the same information but from only the accelerometer.

References

http://ieeexplore.ieee.org/document/7881728/figures

http://s2is.org/ICST-2014/papers/1569981827.pdf

http://papers.www2017.com.au.s3-website-ap-southeast-2.amazonaws.com/proceedings/p351.pdf

http://rsif.royalsocietypublishing.org/content/8/65/1682.figures-only

http://ieeexplore.ieee.org/document/7859881/

https://arxiv.org/pdf/1609.03323.pdf

http://ieeexplore.ieee.org/document/6036745/

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4381862/

Built With

Share this project:

Updates