The Whole Package
The Person Sensor includes a camera and microcontroller in a tiny package, and comes pre-loaded with an ML algorithm that can detect people.
Running machine learning algorithms on tiny edge computing devices is revolutionizing what is possible with our electronic gadgets. Recognizing a person, automatically calling emergency services after a car crash, or detecting when an elderly person has fallen down are now common functionalities built into devices powered by microcontrollers. But these technologies are not being used to their full advantage due to the complexities surrounding building, training, and deploying a machine learning model. The thought of tensors, backpropagation, and loss functions is enough to make most people's heads spin. Fortunately, there are a number of ways to deal with this complexity and make these techniques more accessible to all.
One approach that is increasingly being explored is building the algorithms directly into a sensor. We recently reported on one such effort, and now there is another that has recently been announced. This more recent work is notable for both the simplicity of the technology, as well as the very low price point that makes smart sensor technology accessible even to the hobbyist in single unit quantities. Developed by Pete Warden's new startup Useful Sensors, the Person Sensor is available for $10 and sports an image sensor and microcontroller on the same tiny circuit board. And as the name implies, it runs a machine learning algorithm that can recognize people right out of the box — no need to build or train a model yourself.
It does not take any technical wizardry to use this pre-programmed sensor. Just power it up and point the camera at whatever you are interested in. Results are transmitted over I2C via a Qwiic interface. This information will tell you how many faces have been detected, and where they are relative to the device. There is also a special pin that is used to indicate if any people have been detected in the current image frame. You can even use the Person Sensor to recognize specific faces. This trend towards simplicity in machine learning is reminiscent of the point-and-click interface that made personal computers easy to use for the non-technical person. Maybe point-and-learn is an appropriately pithy way to describe these new sensors? Well, if that ever catches on, remember that you heard it here first at Hackster News (but we are not holding our breath).
As SparkFun points out, the Person Sensor would be easy to incorporate into larger devices that wake up as a person approaches, that mute a microphone when no one is present, or that customize experiences for each user. There are several tutorials available to help new users integrate the Person Sensor into their own projects, whether they want to auto-lock their laptop screen, or integrate the sensor into their own builds using a Raspberry Pi, Arduino, the Raspberry Pi Pico, or Circuit Python. Be sure to also take a look at the Developer Guide for all the info you might want about the Person Sensor.
Taking into consideration the low cost, simplicity, and energy efficiency (approximately 150 milliwatts), we expect to see Person Sensor projects gracing the pages of Hackster.io very soon. If you are in the market for a new gadget to experiment with, you might just want to grab one before they are gone.