 Thanks to artificial intelligence, devices and tools are getting smarter and are getting enhanced with capabilities to provide customized services and make the users life easier. By detecting who is the user of the equipment, devices can customize their behavior to fit user preferences. For example, a user where HVAC can set the thermostat according to user preference, smart coffee machines can prepare your favorite beverage automatically, and even your smart toastel can customize temperature to toast the bread the way you like it. Other use cases are to automate elevators in office buildings or hotels to automatically select the right floor for each person. Or, for example, equipment and machines can adapt their display based on users' registered settings or set height and other ergonomic settings based on users' registered morphology. These are only a few examples among many possible applications. Face recognition is an effective way to personalize smart devices' behavior, display or settings to the user. This advanced functionality is still costly and intrusive when implemented in the cloud and privacy is a strong concern for wider acceptance. To overcome some of these issues, we developed a solution running locally at the edge on a cost-effective microcontroller which can be integrated easily into a large range of equipment. The data remains totally private to the user's environment on the autonomous edge device. Thanks to the powerful STM32 capabilities, STMicroadronics has developed a solution to embed and run face recognition locally on a single cost-effective microcontroller. This is a solution addressing privacy concerns by processing images locally without using the cloud. Implementing face recognition on STM32 MCUs reduces considerably the bomb cost and allows to embed advanced new features into a wide range of devices. To reduce the development effort and to allow to integrate such features easily on your product, we are providing three code examples with our face recognition library in our new function bank. The fp-ai-facereg1 is optimized for the STM32 h747 discovery kit with add-on camera board using our well-known STM32cube.ai tool set. Now let's see the board and the function pack in action and let me show you its capabilities. When the person approaches the camera, the face is detected and displayed in a red bounding box, which means it is not recognized. Now let's press the blue button and register myself. As you can see, I am enrolled as user zero indicated by a green bounding box. My face is tracked in real time and it is even working if my head is tilted or not fully facing the camera like in my registration picture. We are reaching up to four frames per second for the face detection and recognition processing. Now let's see what happens with a different person. This person is not recognized as registered user indicated by the red box. But we can choose to learn new faces directly on the target by pressing the blue button. Let's do that. The bounding box is turning green and the person is now recognized as user one. On the display you can also see the match probability. This solution is ready to be integrated into our STM32 microcontrollers with ARM Cortex M4, M33 or M7 core and a camera module. The code can be extended further by adding IR or time of light sensors to make it more robust to spoofing. Such sensors allow to detect when people for example try to fake identification such as putting a 2D picture in front of the camera. The FP-AI-FaceRack1 function pack will be available in the first half of this year for free. To download from our website, please visit st.com-stm32cube.ai to get access to the stm32cube.ai The wiki and documentation are simply to learn more about our artificial intelligence solutions for microcontrollers. For further discussions, contact us at edge.ai.st.com or via our sales representative.