Unfortunately the default orientation of this screen means that the Pi power cable is inserted from below - and this means that the unit cannot stand up on its own. Final steps - a Caseįor this project we used the HyperPixel 4.0 Square screen - plugged directly into the Pi GPIO bus. This could be the camera, the lack of processing power on the Pi4, the room lighting, camera settings or the faces used (!) - a bit frustrating in the end. It should be noted that in practice it seemed very hard to make the model recognise any emotions other than 'neutral' and 'happy'. The AI2GO facial expression library is capable of detecting a range of expressions, including 😄'happy', 😞'sad', 😠'anger', 😨'fear', 😝'disgust', 😲'surprise', and 😑'neutral'. The code needs some further improvements to handle how it reacts when no faces are detected - an image 'face_makeaface.png' was added to show an invitation to people passing to make a face at the camera. The code was adapted to have an unobtrusive button to do this added onscreen. In running unattended without an attached keyboard/mouse, it was also found to be necessary to add an onscreen 'halt' button in the Python code to shut the system down (just turning off a Pi at the power is not a good idea). > nano /home/pi/.config/autostart/facial_sktopĮxec=/usr/bin/python3 /home/pi/facial_expression.py The file 'xnor-sdk-v1.1.0-rpi3.zip' was copied to the Pi using FileZilla (see below) then, on the Pi: A test mode is also included 'FAKE_DATA = True/False' to generate fake data for testing purposes. The source code 'facial_expression.py' runs the face detection tool, gathers the results from persons passing the camera and selects an appropriate image to show on the screen. Specific AI detection models can be downloaded - in fact we used the libraries held in the SDK. Running on the Pi in Python 3, the following software libraries are used:Īn account was registered on the Xnor site, and the SDK file 'xnor-sdk-v1.1.0-rpi3.zip' was downloaded for the Raspberry Pi (Xnor note this is for the Raspberry Pi model 3 - we were using a Pi 4). HyperPixel 4.0 Square - Hi-Res Display for Raspberry Pi – Touch.The project uses the Xnor AI2GO libraries, and is based on Matt Welsh's example code at Hardware As the camera detects faces with differing expressions, the image onscreen changes accordingly. This project is to develop a Raspberry Pi demonstrator using the Xnor AI2GO library to allow facial expression detection from an attached Pi Camera, and for it to run unattended (no attached keyboard/mouse), changing a 'face' graphic on an attached Raspberry Pi 'Pi Moroni HyperPixel' display. Facial expression detection on Raspberry Pi with attached Pi camera, using the Xnor AI2GO libraries.

Xnor expression