The goals / steps of this project are the following:

1. Load the data set (see below for links to the project data set)
2. Explore, summarize and visualize the data set
3. Design, train and test a model architecture
4. Use the model to make predictions on new images
5. Analyze the softmax probabilities of the new images

Dataset

The dataset used for this project is German Traffic Sign dataset provied by Institut für Neuroinformatik The relevant information about the dataset can be found in the following link: http://benchmark.ini.rub.de/?section=gtsrb&subsection=dataset#Overview

Data Set Summary

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

• The size of training set is 34799
• The size of test set is 12630
• The shape of a traffic sign image is (32, 32, 3)
• The number of unique classes/labels in the data set is 43

Exploratory visualization of the dataset

Here is an exploratory visualization of the data set. It pulls in a random set of 12 images and labels them with the correct names in reference with the csv file to their respective id's.

After this, the dataset structure is analyzed by plotting the occurrence of each image class to get an idea of how the data is distributed. This helps in understanding the data distribution for each class and which class lacks the data.

The data distribution of training data is shown below:

The data distribution of test data is shown below:

The data distribution of validation data is shown below:

The next step is to convert the images to grayscale because the excess information only adds extra confusion into the learning process. The gray scale image is shown below:

The grayscale image is then normalized as it helps in speeding up the training and performance metrics like resources. The normalized images is shown below in comparison with original image

As discussed earlier, the data distibution is not uniform. To mitigate this problem, data augmentation technique is used. Here in this project, data is augmented using some computer vision techniques. The randomized modifications are random alterations such as opencv affine and rotation.

Now, the data distribution is checked by plotting the the training, test and validation data. After augmentation, the number of data in each class is uniform.

The data distribution of training data after augmentation is shown below:

The data distribution of test data after augmentation is shown below:

The data distribution of validation data after augmentation is shown below:

Model Architecture

The model used in this project is LeNet. I referred the foolowing paper for the implementation of LeNet Architetcture. http://yann.lecun.com/exdb/publis/pdf/lecun-01a.pdf The details about the model architecture is shown in the table below.

Layer	Description
Input	32x32x1 grayscale image
Convolution 5x5	2x2 stride, valid padding, outputs 28x28x6
RELU
Max pooling	2x2 stride, outputs 14x14x6
Convolution 5x5	2x2 stride, valid padding, outputs 10x10x16
RELU
Max pooling	2x2 stride, outputs 5x5x16
Convolution 1x1	2x2 stride, valid padding, outputs 1x1x412
RELU
Fully connected	input 412, output 122
RELU
Dropout	50% keep
Fully connected	input 122, output 84
RELU
Dropout	50% keep
Fully connected	input 84, output 43

The training parameters of the model can be seen in the code.

The final model results were:

• Training set accuracy of 100.0%
• Validation set accuracy of 99.1%
• Test set accuracy of 94.1%

Prediction

To check the accuracy of the model, 6 random traffic sign images were chosen and used the saved model weights to classify the images. The results were 100% accurate and the result is shown below.

Reference Links

1. https://arxiv.org/pdf/1409.1556.pdf
2. https://arxiv.org/pdf/1409.4842.pdf
3. https://arxiv.org/abs/1812.03079
4. https://arxiv.org/abs/1512.02325
5. https://developer.nvidia.com/cudnn
6. https://medium.com/@karpathy/yes-you-should-understand-backprop-e2f06eab496b#.vt3ax2kg9