In this project we are exploring the practical use of Edge AI in low-cost distributed home security systems for emerging markets.
A distributed network of sensors of various kinds appropriate for various security use-cases connected over a secure private wi-fi to an aggregation and decision gateway is a simple and well tested architecture. Things start becoming interesting when the distributed sensor systems use previously trained inferencing models to locally detect patterns in the time series of its sensor data at real-time speed. Receiving semantically richer aggregated events information instead of unclassified raw sensor data streams of various kinds enables the aggregator station to make higher level decisions fast enough for initiating more precise and responsive event handling actions.
This project belongs to the category of hands-on projects that we deal with in the upcoming upcoming E-Vidya Workshop on Embedded Edge AI.
A key aspect of this project is the choice of hardware for the various types of system nodes that need integrated low-power wi-fi as well as enough memory and processing power for real-time inferencing on specialized sensor data streams or event data streams. The new BeagleBone AI seems like an interesting candidate for the aggregation station.
Initial experiments show promising prospects. More details will come as we make further progress.