Traffic Sign Recognition

---

---

###Data Set Summary & Exploration

####1. Dataset: German Traffic Sign Dataset

• The size of training set is 34799
• The size of the validation set is 4410
• The size of test set is 12630
• The shape of a traffic sign image is originally (32, 32, 3). (32, 32, 1) Gray.
• The number of unique classes/labels in the data set is 43

####2. Distribution of training data

###Design and Test a Model Architecture

####1. Images are converted to grayscale. Because color does not help to differentiate the traffic sign.

After that operation pixel values are normalized ((X/255)-0.5) because.

Possible improvements: Generating artificial image and adding some noise.

####2. Final model architecture

My final model consisted of the following layers:

Layer	Description
Input	32x32x1 Gray image
Convolution 3x3	1x1 stride, valid padding, outputs 30x30x8
ELU
Max pooling	2x2 stride, outputs 15x15x8
Convolution 1x1	1x1 stride, outputs 15x15x8
Convolution 3x3	1x1 stride, valid padding, outputs 13x13x16
ELU
Max pooling	2x2 stride, outputs 6x6x16
Convolution 1x1	1x1 stride, outputs 6x6x16
Fully connected	6x6x16 = 576
Fully connected	120
Dropout	0.6
Fully connected	84
Dropout	0.6
Fully connected	43
Softmax

To train the model, I used TensorFlow AdamOptimizer with learning rate 0.001, 20 Epoch , 128 batch size

It is bacisly LenNet architecture. I reduced filter sizes from 5x5 to 3x3 and I added two 1x1 convolution layers and dropouts.

Basic LeNet-5 results are good for classify traffic signs: (learning rate 0.001, 20 Epoch , 128 batch size)

• training set accuracy of %99.6
• validation set accuracy of %93.0
• test set accuracy of %92.2

My final model results were:

• training set accuracy of %99.8
• validation set accuracy of %94.5
• test set accuracy of %92.4 Changes have allowed us to get better results.

I also designed a inseption module you can find it in the code but, I could not test it.

###Test a Model on New Images

####1. Choose five German traffic signs found on the web and provide them in the report. For each image, discuss what quality or qualities might be difficult to classify.

Here are five German traffic signs that I found on the web:

####2.

Here are the results of the prediction:

Image	Prediction
Bumpy road	Bumpy road
Stop	Stop
No entry	No entry
No passing	No passing
Priority road	Priority road
Ice/snow	End of no passing by vehicles over 3.5 metric tons
Children crossing	Right-of-way at the next intersection

####3. The top 5 softmax probabilities for each image along with the sign type of each probability.

Probability	Prediction
%99	Bumpy road
%99	Stop
%100	No entry
%100	No passing
%100	Priority road
%82	End of no passing by vehicles over 3.5 metric tons
%88	Right-of-way at the next intersection

When we want to classify clearly visible traffic signs, the outcome is very likely to predict correctly. But weather conditions such as snow and fog reduce the success rate and cause misclassification.