wonnor-pro/MTCNN-pi

Run MTCNN on Raspberry Pi

MTCNN-pi

We aim to implement the real-time face recognition based on Multi-task Convolution Neuron Network on Raspberry Pi 3B.

This project was developed by our develop team based on

• Previous work from the Github Project
• Original Paper: Joint Face Detection and Alignment using Multi-task Cascaded Convolutional Networks (K. Zhang, Z. Zhang, Z. Li, Y. Qiao, 2016).

Credit to our team members:

• Jinghong Chen, @EriChen0615 on Github; jc2124@cam.ac.uk
• Xiaoqiao Hu, xh297@cam.ac.uk
• Connor Wang, @wonnor-pro on Github; xw345@cam.ac.uk

To understand this Neuron Network, you can refer to my post Understand MTCNN (Multi-task Cascaded Neuron Network) in 10min

---

Hardware

• Raspberry Pi 3B (Operating system: Raspbian)
• Camera Module

Software

• Python 3.5.3

Settings

Click the link to see the installastion instruction.

1. Environment Setup and Dependencies

• Install dependencies
• Check Python version and install pip3
• Install Protobuf 2.6.1
• Install OpenCV 3.3.0
• Install caffe

2. Install the Camera Module

• Install the camera hardware
• Configure the Raspberry Pi
• Test the camera module

File Structrues

You should be able to find three demo files in this repo as we use different strategies to speed up the process. It is noteworthy that the accuracy varies from method to method.

• demo_slow.py
• demo_scales.py
• demo_multiprocess.py (Do not run the multiprocess demo for a long time on your Raspberry Pi, as it fully uses its cores and may overheat the chip.)

As the file name suggestes, we used methods like grayscale/selected scales/multiprocessing to improve the inference speed for real-time face detection.

Requisitions regarding reposting please contact wonnor.cam@gmail.com.