In this project, I used LeNet to classify traffic signs. By using the using the German Traffic Sign Dataset, I got a validation accuracy of 0.990 and a test accuracy of 0.930.
After the model is trained, I tested my model program on five new images of traffic signs downloaded from the web. The result showed that my model just predicted two pictures correctly. The model still need to be improved.
There are two ways I think can be improved. First, take the channels of color into account. The input image is 32321. The low resolution make outline of images blurred, so the channels of color can provide mor information to model. Second, we can generate more training data by jittering and transforming source training images.
This project requires Python 3.5 and the following Python libraries installed:
- Jupyter
- NumPy
- SciPy
- scikit-learn
- TensorFlow
- Matplotlib
- Pandas (Optional)
Run this command at the terminal prompt to install OpenCV. Useful for image processing:
conda install -c https://conda.anaconda.org/menpo opencv3
- Download the dataset. This is a pickled dataset in which we've already resized the images to 32x32.
- Clone the project and start the notebook.
git clone https://github.com/udacity/CarND-Traffic-Signs
cd CarND-Traffic-Signs
jupyter notebook Traffic_Signs_Recognition.ipynb
- Follow the instructions in the
Traffic_Signs_Recognition.ipynbnotebook.