COVID19 isolated at home many of us, including our elderly parents and grandparents. Not being able to check on them regularly elevates the risks that they are exposed to such as falls, gas leaks, flooding, fire and others.
What it does
Ambianic.ai is an end-to-end Open Source Ambient Intelligence project that removes the stigma associated with surveillance systems by implementing privacy preserving algorithms in three critical layers:
Peer-to-Peer Remote access Local device AI inference and training Local data storage Ambianic.ai observes a target environment and alerts users for events of interest. Data us only available to homeowners and their family. User data is never sent to any third party cloud servers.
Here is a blog post that goes into the reasons why we started this project: https://blog.ambianic.ai/2020/02/05/pnp.html
And here is a technical deep dive article published in WebRTCHacks. It clarifies that it is absolutely possible to build a privacy preserving surveillance system, despite popular cloud vendors making us believe that all user data belongs safely on their cloud servers: https://webrtchacks.com/private-home-surveillance-with-the-webrtc-datachannel/
How we built it
Ambianic.ai has 3 main components:
Challenges we ran into
Challenges include selecting high performance, high accuracy and low latency AI models to detect events of interest on resource constraint edge devices.
Another challenge is taking into account user local data to fine tune AI models. Pre-trained models can perform reasonably well, but they can be improved with privacy preserving federated learning on unique new local data.
Accomplishments that we're proud of
Ambianic.ai has been in public Beta for several weeks helping a number of users in their daily lives. Some users report success in keeping an eye on their elderly family members: https://twitter.com/mchapman671/status/1230931722650423299
What we learned
Although the project sets ambitious goals, there seem to be sufficient enabling Open Source frameworks and community momentum to drive the ongoing success.
What's next for Remote Elderly Home Care via Privacy Preserving Surveillance
We need to work on these major areas:
Recruit volunteers in the home care community to test the system and provide feedback Select more models to address open use cases such as fall detection, gas leaks and others Work on implementing Federated Learning infrastructure to fine tune initial pre-trained models.