training loss tends to zero!
We find it necessary for busy people like university students to keep track of their life with ease. If a device can automatically record users' life pattern, users should, therefore, have a deeper understanding of how to organize their life and lead to far higher efficiency.
What it does
Our project is an Arm Mbed attached to user's right forearm, which is able to automatically classify three different kinds of users' activity including working, walking and resting. Moreover, the project can send email to users' mailbox when any abnormality is detected around users through the air quality and ambient light sensor on Mbed. Last but not least, we also implemented a "raise to show" module which will only display info including time and sensor readings to users when users raise their hand towards their faces.
How I built it
We built it under the online tutorial of Mbed, which helped us learn to utilize MQTT to achieve high rate data online data transfer and the sensor APIs. The classifiers of three activities were achieved with accelerometer and RNN(using LSTM and 1D Conv layer), implemented with keras.
Challenges I ran into
Collecting data and train a decent model within 24h is really challenging, and using MQTT server which we have never come across were really challenging
Accomplishments that I'm proud of
We finished most of the functionalities we expected, which is more them amazing considering the short time we are given.
What I learned
Everything is possible! And plan carefully ahead!
What's next for Arm_Hackathon_Wearable_Mbed
More activities can be classified, and user-customized unsupervised learning is also possible