It is a final proejct for embeded system in 2023. The normal kiosk is designed for adults. It is difficult to use kiosks for people with physical difficulties. So, i suggested kiosk arm which tracking user's face to improve accesibility.
- raspberry pi 4B
- Google coral TPU
- Camera
- Servo motor
- tact switch
To recognize face region, I used Goolge coral accelerator to avoid delay. It was hard to run large deep-learning model on raspberry pi. So, this TPU computes face recognition deep-learning model. SSD MobileNet(Pre-trained) was used to do this.
After get face region from TPU, compute the error between ceter pixel of image and center pixel of face. Then, apply PID control to get each servo motor position.
Also, there is 2-interrupt service routine(ISR). Not pooling
- Reset ISR -> user can initialize servo motor position by push tact switch 1.
- Shutdown ISR -> user can shutdown main program by push tact switch 2.
https://youtu.be/_tYapHzxim4 You can check the real-time demonstration of our kiosk arm via above youtube link.