Traffic Sign Recognition

Overview

In this project,I used deep neural networks and convolutional neural networks to classify traffic signs. I train and validate a model so it can classify traffic sign images using the German Traffic Sign Dataset. After the model is trained, I then tested the model on images of German traffic signs that you find on the web.

Steps

The goals / steps of this project are the following:

• Load the data set
• Explore, summarize and visualize the data set
• Design, train and test a model architecture
• Use the model to make predictions on new images
• Analyze the softmax probabilities of the new images

Instructions

• Open Traffic_Sign_Classifier.ipynb and Traffic_Sign_Classifier.html for code
• Open report.md and for work summary

Writeup

---

Build a Traffic Sign Recognition Project

The goals / steps of this project are the following:

• Load the data set (see below for links to the project data set)
• Explore, summarize and visualize the data set
• Design, train and test a model architecture
• Use the model to make predictions on new images
• Analyze the softmax probabilities of the new images
• Summarize the results with a written report

---

Data Set Summary & Exploration

I used the pandas library to calculate summary statistics of the traffic signs data set:

• The size of training set is 34799 image
• The size of the validation set is 4410 image
• The size of test set is 12630 image
• The shape of a traffic sign image is 32x32
• The number of unique classes/labels in the data set is 43 classes

2. Exploratory visualization of the dataset.

Here is an exploratory visualization of the data set. It is a bar chart showing how the data ...

Design and Test a Model Architecture

Data Preprocessing

1. Normalizing the data using the OpenCV Library Min-Max Normalization:

   • Reduced the pixels values between -1. and 1. to centralize the data around the origin
   • Reduced the data's mean and standard deviation for better and more valuable training.

   Here is an example of a traffic sign image before pre processing.

   

   Here is an example of a traffic sign image after pre processing.

   

2. Shuffling the entire data set, and labels, using the sklearn library:

   • Randomly shuffling the data for training to attain a random distribution throughout each batch for Stochastic Gradient Descent.
   • The shuffling had a major role in the training of the model and a huge impact on the network's accuracy.

Model Architecture

This model architecture follows the implementation of the LeNet CNN.

My final model consisted of the following layers:

Layer	Description
Input	32x32x3 -- RGB image
Convolution 5x5	1x1 stride + VALID padding + outputs 28x28x6
RELU
Max pooling	2x2 stride + outputs 14x14x6
Convolution 5x5	1x1 stride + VALID padding + outputs 10x10x16
RELU
Max pooling	2x2 stride + outputs 5x5x16
Fully connected	120 unit
RELU
Dropout	0.5 keep_prob
Fully connected	83 unit
RELU
Dropout	0.5 keep_prob
Output	43 Logits

Model Training

Tuned Parameters

The choice of these hyperparameters required several trials to reach this final combination which represents the maximum performance achieved.

1. Epochs = 40
   Chosen upon the fact that the model reaches a plateau by the 40th epoch

2. Batch Size = 64
   Appropriate and efficient batch size

3. Learning Rate = 0.001 Through many tests, this learning rate was right before overshooting, nonetheless fast and converges

4. Mean = 0 & Standard Deviation = 0.1 Values fed for the tf.truncated_normal() function for weight initialization

5. Dropout = 0.5 The probability for the dropout layers which decreased vastly the overfitting of the dataset

   • Used the tf.nn.softmax_cross_entropy_with_logits() function to calculate the logits probabilities using: softmax + the cross entropy
   • Used the Adam Optimizer for training the network with backpropagation and stochastic gradient descent.

Model Performance

My final model results were:

• Validation Set Accuracy = 96.07 %
• Test Set Accuracy = 94.27 %

Testing the Model on New Images

Introduction to the Chosen Images

These are 7 German traffic signs that I found on the web:

After pre processing

Prediction Results

The model was able to correctly guess 6 of the 7 traffic signs, which gives an accuracy of 85.714 %. This compares favorably to the accuracy on the test set of 94.2 %

This is a table representing the 7 images tested, and the top 5 softmax probabilities produced by the network analyzing these images:

Feature Map Visualization of 2nd Convolutional Layer

The following image, representing a traffic sign is run through the network.

This image activates 16 different Feature Maps in the 2nd Convolutional Layer, representing what did each feature map observe in this image. as visualized below: