Video camera are expensive and cannot see and report every crime. A cheap portable recorder would allow us to detect more crimes and report them to the police. It could be deployed in poor or remote area where cameras are etheir not practical or expensive.
What it does
It is a small device that records and analizes audio continuously for threat and other events. Once it detect an event, it will trigger an allert to the appropriate person.
How we built it
We first build an Android app and try to build a independent IoT device. They record audio continuously and send it to a cloud server that processes the information. The result is feed to an A.I. model who evaluates if the audio contains an important event (gunshots). The response is sent via notification to phones are possibly other devices.
Challenges we ran into
Lack of available and compatible hardware. Poor quality datasets. Many moving parts that had to come together. Audio file encoding.
Accomplishments that we are proud of
Traning a CNN A.I. model with 70% accuracy in a 24h hackathon. Creating and deploying an Android app with a backend and a notification system. Utilizing Docker and AWS to host our service on the cloud. Working with IoT devices.
What we learned
That microphones are not created equal. Make sure we have all the data available before starting to train a model. Audio encoding is a very capricious thing.
What's next for MobileEAR
A better A.I. model. Complete the integration with the IoT side. Fix some bugs in the interaction between the backend and the A.I. model.