Third Place Winner:
SafeHaven: Real-Time Reassurance. Re:invented.
SafeHaven was designed to protect vulnerable people living alone, by enabling them to identify "Who Is At The Door" using an Alexa Skill. Unknown visitors trigger SMS or email alerts to relatives or carers, via an SNS subscription.
What it does
DeepLens acts as a sentry on the doorstep, storing the faces of every visitor. When a visitor is "Rekognised", their name is stored in a DynamoDB table, ready to be retrieved by an Alexa Skill.
Created By: Nathan Stone and Peter McLean
Learn more about Nathan, Peter and the SafeHaven project in this AWS Machine Learning blog post.
How we built it
AWS have democratised many complex computational elements (particularly in the realm of machine learning), making it possible for developers to build complex systems by configuration.
SafeHaven is only possible because AWS DeepLens allows us to deploy a light-weight, yet powerful Face-Detection neural network to a compact device. DeepLens minimizes network traffic, so we don't need to stream high volumes of video data to the cloud.
What's next for SafeHaven
We can see the system scaling up to be used in multi-tenant assisted care facilities, where the door can only be opened to recognised visitors, even when vulnerable or elderly residents may have been persuaded to permit entry. In this case, on-site supervisors will receive alerts so they can challenge unknown visitors.
We think the system could be enhanced with Echo Show, which would display the visitor’s face along with a caption to describe who they are.
We believe that the AWS eco-system would allow us to offer the system to multiple DeepLens owners, each with their own database of trusted visitors.
We would extend the existing SMS with a “Trust This Face” API Gateway URL, which would add the face to the list of known visitors, making it easy for guardians to maintain the underlying database remotely.
We might change the model from Face-Detection to Object-Detection, to identify the objects that people are holding – perhaps enabling the system to identify that the visitor is a postman, even though their face is not recognized.