Face Recognition

Our smart home has a growing number of cameras and they are all under it's control, so that it can change the way they behave based on 'whole home context'. This means it is also aware of all the images and video that has been captured and can process it in near real-time or post-process it at a later date.

Real-time processing enables alerts to be generated based on recognised or unrecognised faces being spotted. Our smart home also has the ability to send captured footage or a link to it in our cloud storage.

Post-processing allows our smart home to learn new faces and people. It also allows it to provide an activity summary or report.

Design Thinking

We don't want face recognition to be a function of individual cameras or 'smart' devices in the home. It is much better to enable face recognition with full smart home context, to then know what best to do with the outputs. By making it a generic capability of our smart home, it means that any camera we add can inherit this capability, regardless of the make/model, cost, etc. Basically, any cheap camera can have face recognition capability using our approach and we retain full control over how this happens, when this happens and any actions arising from it.

We can do this because we retain control of all footage captured and it is stored locally within our smart home. This also means we are not tied to any specific cloud storage service and our smart home can choose to synchronise some or all of the captured footage to our chosen cloud storage service if required. This is our preferred method to enable all family members with teh means to view and review captured media. We have several other methods to enable this though.

Not only does this approach provide the best performance and the best user experience, but it also provides a much wider set of smart home capabilities to our family. Some examples:

  • All family members (with permissions) can view, review and delete media.
  • We could ask our smart home to "send me pictures of all visitors".
  • Our smart home can learn about regular callers, such as the postman and window cleaner. It could then provide facial images (using face extraction software) of all unknown people that called at our home in the last day, week, month, etc.
  • We can currently ask our smart home to let us known when family members leave home or get home but this would allow us to extend this capability to any known/recognised person.
  • "Let me know if their is an unknown person in the garden"

Future Thoughts

Once we have completed this project, then we plan to look at more generic object recognition algorithms, inlcuding license/number plate recognition as a generic capability for all our smart home's IP cameras.

Share ...
We are on ...
Facebook Twitter
YouTube Flickr Follow us on Pinterest