traffic_sign_recognition

Traffic sign recognition based on Dlib and OpenCV project

School project (POV)

Learning and detecting traffic sign in images and video.

Dependency

• Dlib 19.2 (need compile and copy to resources/dlib-19.2 see README.md)
• OpenCV 3.1.0 (need compile and copy to resources/opencv-310 see README.md)
• libjpeg
• libpng
• zlib

Use

After build this project in VS2015 you can find all necessary files in build folder

Folder structure:

• build
  • signs - folder with images of signs
  • datasets - contains xml documents, input floder for trainer
  • descriptor - contains descriptors, output folder for trainer
  • tools - contains imglab.exe for creating datasets

trainer run parameters

• without parameters - learn all datasets in folder datasets
• one of these: "stop", "crossing", "give-way", "oneway-exit" , "main", "parking" - learn only specified dataset. see ./global/config.h

run example:

./traffic_sign_trainer
./traffic_sign_trainer stop
./traffic_sign_trainer main

detector run parameters

• "-i image.bmp" detect signs in image.bmp
• "-v video.mp4" detect signs in video.mp4

run example:

./traffic_sign_detector -i image.bmp
./traffic_sign_detector -v video.mp4

Image sets

Images are mainly from Google street view. Some images are from searching at Google images.

Need improvement

• better datasets (mainly for stop sign)
• add negative and test datasets
• for high videoresolution images and videos:
  • specify detection area - blue color => blue signs in limited detection area
  • upsample video frames for better recognition
  • perfomance => speedup