Traffic-sign-detection

Project of photoelectric information processing experiment in ZJU, ISEE

Overview

This is a small embedded project based on Raspberry Pi. The 'sparrow' may be small, fully-equipped.

It involves the following:

• Colour extraction and ellipse detection based on traditional methods
• Video and order transfer between Raspberry Pi and PC based on socket programming using TCP/UDP protocol
• Simple remote controller based on PyQt5
• Image multi-classification problem based on classical neural network (lenet here)
• Application of SSD pedestrian detection model
• Raspberry Pi level output based on RPi.GPIO library
• Raspberry Pi built-in audio player driver calls

More details in our project report (in Chinese) and demo video.

References

• Ellipse detection
• SSD model
• Image classifer framework
• Driver board datasheeet

File structure

Run demo

• on your PC

  • get your LAN IP address of PC

     ipconfig
    

  • run on your cmd or Anaconda Prompt

    python3 PC.py --host=your_IP_address 
    

  • if you want to use TCP protocol to transfer video (not recommend)

    python3 PC.py --udp=False --host=your_IP_address 
    

• on your Pi

  • get your LAN IP address of Pi

     ifconfig
    

  • run on your terminal

    python3 Pi.py --host=your_IP_address 
    

  • if you want to use TCP protocol to transfer video (not recommend)

    should be same as the protocol you chose on your PC

    python3 Pi.py --udp=False --host=your_IP_address 
    

  • if you want complete all calculations on your Pi (not recommend)

    python3 run_direct.py
    

Tools

• color_extract.py

  It is a small interactive UI for easy access to get color data in video

• remote_controller.py

  It is a simple UI which can control your car

• rename.py

  It is a script which can rename all your files in one dir

• utils.py

  It contains some useful functions

• grasp.py

  It can determine the flag type in an image or video and crop it out for use in data set construction and early debugging

Learning-related

• lenet.py

  It is the network struct of lenet

  

• train.py

  Train the model of lenet, which implements a seven-classify (one background, six signs) task

  

  

• predict.py

  Predict the label of single image

• pedestrian.py

  It is a script that calls the SSD model