Smeezys: Smart Yeezys

• 

  Shoe as sensors are being attached

• 

  Shoe with sensors fullt attached

• 

  Nearly Complete Shoe

• 

  Shoe on Foot Model

• 

  A Comparison of the accuracies of various machine learning algorithms in classifiying various activities

• 

  Example Confusion Matrix 1 (High Accuracy)

• 

  Example Confusion Matrix 2 (Low Accuracy)

• 

  Real Time heat map for individual sensors in grafana

Inspiration

Kanye West's Yeezy's have pushed the bounds of footwear aesthetics; we want to do the same with footwear analytics. Like Kanye, we are pushing boundaries farther than they need to be pushed.

What it does

Smeezys use a custom designed comparative pressure sensor system to accurately measure the movement and strain on a user's foot, which is then processed and served through our suite of data visualization tools.This technology uses machine learning and convolutional filters to accurately determine what the user is doing (99.17% +/- 0.87%), generate real-time pressure heatmaps, and give cues to improve running/walking form. These technologies can improve athletic performance and allow disabled individuals more freedom of movement.

How we built it

Our custom built comparative pressure sensors use the change in the resistance of conductive foam when compressed to measure pressure, and serve as part of the in-sole in our shoe design. The input of those sensors is aggregated by an arduino microcontroller and uploaded to an alpine docker database using influxdb. These data are used by our machine learning and data visualization engines, built in python, C, Grafana, and HTML5.

Challenges we ran into

The hardware component of our build was the most compelling and difficult part of the hack. Designing and fabricating the sensors, disassembling a shoe, integrating the sensors and arduino, managing wiring, reassembling the shoe, and compiling activity data for our shoe model took hours upon hours and challenged our electrical, materials engineers, and professional foot model.

Accomplishments that we're proud of

We designed and built our own sensors because nothing on th market meets current needs. Our machine learning is amazingly accurate. (Credit to convolution and our feature generation engine.) We did our own low-level optimizations for visualization in C. We have the flyest shoe game.

What we learned

We learned a lot about mechanical engineering, systems design, and visualization tools for large datasets.

What's next for Smeezies: Smart Yeezies

We would like to expand the technology to create a consumer grade product, especially focusing on data analytics for professional athletes, and custom insoles + an AI assistant aimed at personal physical training for the disabled.

Built With

• arduino
• bash
• c
• duct-tape
• influxdb
• pandas
• python
• scikit-learn
• zigbee-(802.15.4)