Udacity SDCND Project_3

Traffic_Sign_Classifier

In this project, I use a LeNet-5 convolutional neural networks to implement a classifier to distinguish German traffic sign. LeNet-5 is the most basic structure of the CNN.

After the conv layer and the pooling layer are repeated twice, fully connected connection layer is used three times.

---

Process

• Load The Data

In the given file, load the data in the following format

Number of training examples = 34799

Number of validation example 4410

Number of testing examples = 12630

Image data shape = (32, 32, 3)

Number of classes = 43

• Dataset Summary & Exploratory Visualization

visualize an image about each label and represent the ratio of the element number of the each class to total.

• Preprocess Data

Convert the image to gray color and normalize it for better learning.

• Design LeNet-5

By using tensorflow, conv layer, pooling layer, activation, and fully connected layer are implemented. Then compares the calculated logit with the actual label and optimizes it to minimize the difference.

• Train the model and validate & test it (Our goal is to achieve validation accuracy of over 93%)

The model is trained by passing the input sign through a convolution neural network and minimizing the difference between the predicted value and the actual value.

• Test a Model on New Images

Apply the six German traffic signs received from the Internet to the learned models to ensure that the learned models correctly distinguish traffic signs.

---

Configuration

• Traffic_Sign_Classifier.ipynb

contains a complete code to implement the actual project In addition, the results and descriptions of each process have been added to confirm the entire process

• Writeup.md

describes the important process in detail

• Traffic_Sign_Classifier.html

converts ipynb file to html

• New_images

contains new images obtained from internet

• signnames.csv

contains what the German traffic sign label means

• save_train.ckpt

contains parameter values of the trained model

• Reference : https://github.com/netaz/carnd_traffic_sign_classifier.git

  I referenced this link to visualize the data.