In my Mailing List Newsletter Today I got A Post on the Oura Ring and Google Nest Sleep Data does Edge Impulse have a recommended or general approach to inputting and learning from biosensors of this type? Can it stream multiple biosensors for input and classify this time series for a physiological condition or emotion classification?
Hi @laefsky There’s a great paper by Marco Altini on how Oura does sleep stage prediction with their new algorithm. It’s using PPG + temperature + accelerometer data in a single model to predict the sleep stage. All of this is modellable with Edge Impulse. https://www.mdpi.com/1424-8220/21/13/4302
Jan: What is the procedure to recognize multiple real time sensor streams like this in classified categories with continuous streams of data in Edge Impulse? Do you have it available as a sample project?
@IML2 The easiest way is to (during data collection) resample all sensors to the same frequency, and treat it as just one sensor with multiple channels. Then during inference you just recreate this, e.g. see how we feed in multiple sensor axes here https://docs.edgeimpulse.com/docs/cli-data-forwarder#classifying-data-arduino. Then in your impulse you can route the different sensors to different DSP blocks (this will run fine on device).
Is there project documentation which is non-confidential that describes the whole process of using the Oura Ring to provide data, extract features, train and inference as well as any signal processing where the signals from the Oura Ring were used in a real-time classifier for sleep staging? If not what is the closest project documentation which is non-confidential?
@IML2 The closest is the paper linked above ([https://www.mdpi.com/1424-8220/21/13/4302 (https://www.mdpi.com/1424-8220/21/13/4302)).