Real-time traffic-signs recognition using YOLOv3

Bachelor's Thesis

Student: Dương Lữ Điện

Supervisor: Trần Công Án, PhD.

We use YOLO algorithm in our dataset and try to detect some types of traffic signs in Vietnam: 102, 130, 131, 201a, 201b, 202, 203, 205, 207, 208, 209, 221, 224, 225, 233, 245, 302, 303, 423, crowded, end_crowded, traffic_light

Important Notice

This project was published under GNU General Public License v3.0. Make sure that you have read the LICENSE carefully before using it in your project.

Resources

• Report PDF
• Fozen model
• Slides
• Dataset: send an email to duongludien@gm*il.com
• TensorFlow checkpoint: send an email to duongludien@gm*il.com

How to use?

1. Navigate to your local repo.

$ cd YOUR_LOCAL_REPO/yolov3/

2. Create a virtual environment and install requirements

$ virtualenv test_env --python=python3.6
$ source test_env/bin/activate
(test_env)$ pip install -r requirements.txt

3. While requirements are being installed, download frozen model in the Resources section and put it into yolov3 directory.

4. After all finished, run the demonstration

• For predicting images:

(test_env)$ python predict_frozen.py YOUR_IMAGES_DIRECTORY_PATH

Press Left for previous, q for exiting and another key for next

• For predicting videos:

(test_env)$ python predict_video_frozen.py YOUR_VIDEOS_DIRECTORY_PATH

Press q for exiting

Note: If you want to develop your own model from my code, install all requirements yourself :)