We live in a world of data. Virtually everyone talks about data and the potential value that we can extract from it. But raw, massive amounts of data is complex and hard for us to interpret. Machine learning techniques have during the past few years made it possible to understand this data better and be able to leverage upon it. Whereas most of this value so far has been added in online businesses it now also starts to enter the physical world, where the data is generated by sensors.
But the path from sensor data to an embedded AI model seems almost insurmountable. Writing embedded software is known to be very time consuming and take at least 10-20 times longer than desktop software development. In our new white paper we will walk you through a real AI project from data collection to the embedded application and share some important findings that we made along the way.
Today the vast majority of signal processing software that processes and interprets sensor data is based on traditional methods: transforms, filters, statistical analyses etc. It is designed by a human that by using his/her domain knowledge is looking for some particular “fingerprint” in the data. Quite often this fingerprint is a complex combination of events in the data, and machine learning is needed to successfully resolve the problem.
To be able to process the sensor data in real time, the machine learning model needs to run locally on the chip close to the sensor itself - usually called “the edge”. In this paper we will go through how a machine learning application can be created - from the initial data collection to the final embedded application. As an example we will demonstrate a joint project between Imagimob and the radar technology company Acconeer.
Imagimob and Acconeer teamed up in 2019. Both companies focus on providing powerful applications on hardware devices where low power consumption is key. The cooperation therefore represents a good match in terms of creating radically new and creative embedded applications.
Acconeer's A111 Pulsed Coherent Radar (PCR) sensor is optimized for high precision, measurement and detection - with a small footprint and ultra-low power consumption. The XM122 IoT module - where the sensor is integrated on a module optimized for IoT - is the world's first radar system that can run on a coin-cell battery.
The radar signal contains information about relative distance to and speed of reflecting objects and many of radar use cases do not need AI/ML to interpret the data from the radar sensor. But the radar signal is complex to interpret, so for more advanced services, such as detection of complex micro gestures, AI/ML is necessary in order to get good results.
The goal for this project was to create an embedded application that could classify five different hand gestures in real time using radar data. With its small size the radar could be put inside a pair of earphones and the gestures would work as virtual buttons to steer the functionality that is usually programmed into physical buttons. The end product for the project was decided to be a robust live demo at CES 2020 in Las Vegas.
Read the complete story in our White Paper – From data collection to embedded gesture detection library. Fill out the form below to get access.
It’s now possible to include Tensorflow AI models in your C-...