The purpose of this project is to implement deep learning technique to classify the traffic sign image into their relative classes, the datasets that we had utilized is Traffic Sign Recognition Database (TSRD) from Chinese Traffic Sign Database (CTSD), we have applied various data preprocessing technique such as class imbalance handling, normalization etc. We have conducted transfer learning on three models, VGG-16, ResNet-50, and MobileNetV2 and also apply fine-tuning the entire model, only we proceed to evaluate their performance on the test image set. We also implement GUI for demo on traffic sign classification
The TSRD includes 6164 traffic sign images containing 58 sign categories. The images are devided into two sub-database as training database and testing database. The training database includes 4170 images while the testing one contains 1994 images. All images are annotated the four corrdinates of the sign and the category.
Link to TSRD dataset: Traffic Sign Recognition Database
Note: Jupyter Notebook are required for the demo, while if test run our whole pipeline, you need to have python environment that has access to gpu, or you could try run it on Google Colab
- Go to TSRD to download all the dataset and annotation, if only want to try demo, download the test image dataset should be enough
- Download the T2_G2 notebook file to test run our project, otherwise download Demo notebook to just demo on our model
- Remember to download all the models and allocate them in the same path as the notebook
- Enjoy!
| Student Name | Student ID | Role | Email Address |
|---|---|---|---|
| Ng Wei Yu | 2207448 | Leader | weiyung0091@1utar.my |
| Tan Kai Jun | 2206494 | Member | kaijuntan423@1utar.my |
| Ler Jun Wei | 2207200 | Member | luci6n231@1utar.my |
| Wong Kenji | 2206455 | Member | wkenji02@1utar.my |
w[i]_tutorial file or directory are just weekly activites/project for our this course project, can take a look if interested or just ignore it