The goal for this project was to build a traffic sign classifier by training a convolutional neural network to identify German traffic signs from the supplied data set.
The project Jupyter notebook can be found on GitHub.
Three separate data files containing training, validation and test datasets were loaded and un-pickled. The image data and labels were then extracted for each set.
Next the data was analysed to determine the number of samples in each set, the image shape, the number of channels in each image and the total number of classes.
Number of training examples = 34799
Number of validation examples = 4410
Number of testing examples = 12630
Image data shape = (32, 32, 3)
Number of channels = 3
Number of classes = 43
Mapping from Class ID to Sign Name
--------------------------------------------------------------
ClassId SignName
0 0 Speed limit (20km/h)
1 1 Speed limit (30km/h)
2 2 Speed limit (50km/h)
3 3 Speed limit (60km/h)
4 4 Speed limit (70km/h)
5 5 Speed limit (80km/h)
6 6 End of speed limit (80km/h)
7 7 Speed limit (100km/h)
8 8 Speed limit (120km/h)
9 9 No passing
10 10 No passing for vehicles over 3.5 metric tons
11 11 Right-of-way at the next intersection
12 12 Priority road
13 13 Yield
14 14 Stop
15 15 No vehicles
16 16 Vehicles over 3.5 metric tons prohibited
17 17 No entry
18 18 General caution
19 19 Dangerous curve to the left
20 20 Dangerous curve to the right
21 21 Double curve
22 22 Bumpy road
23 23 Slippery road
24 24 Road narrows on the right
25 25 Road work
26 26 Traffic signals
27 27 Pedestrians
28 28 Children crossing
29 29 Bicycles crossing
30 30 Beware of ice/snow
31 31 Wild animals crossing
32 32 End of all speed and passing limits
33 33 Turn right ahead
34 34 Turn left ahead
35 35 Ahead only
36 36 Go straight or right
37 37 Go straight or left
38 38 Keep right
39 39 Keep left
40 40 Roundabout mandatory
41 41 End of no passing
42 42 End of no passing by vehicles over 3.5 metric ...
A graph was plotted showing the number of samples for each class in the training set. From the plot it can be seen that some signs have many more samples than others. The lack of data for certain signs could make it harder to train the network to recognize these signs. If however the number of samples of each sign corresponds to the probability of encountering that sign on the road then the we don't want a more equal sample number as this will not favor the more common signs. The data could be augmented to supply additional training samples to improve the results of the network.
Finally a sample of each image class was plotted with it's corresponding label. From the sample images it can be seen that some images are very dark making it hard to identify the sign. Preprocessing the images by adjusting brightness and contrast could help the network perform better.
The data was preprocessed by normalizing it. The results of the normalisation process on the images can be seen below.
My final model consisted of the following layers:
| Layer | Description |
|---|---|
| Input | 32x32x3 RGB image |
| Convolution 3x3 | 1x1 stride, valid padding, outputs 28x28x6 |
| RELU | |
| Max Pooling | 2x2 stride, valid padding, outputs 14x14x6 |
| Convolution 3x3 | 1x1 stride, valid padding, outputs 10x10x16 |
| RELU | |
| Max Pooling | 2x2 stride, valid padding, outputs 5x5x16 |
| Flatten | outputs 400 |
| Fully Connected | outputs 120 |
| RELU | |
| Fully Connected | outputs 84 |
| RELU | |
| Fully Connected | outputs 10 |
The following parameters where adjusted until an acceptable accuracy was achieved. The learning rate was initially set at 0.005 but after 10 Epochs it was decreased by 20% each Epoch.
| Parameter | Value |
|---|---|
| mu | 0 |
| sigma | 0.1 |
| Batch Size | 250 |
| Ephocs | 30 |
| Learning Rate | 0.005 |
The model achieved an accuracy of 0.996 on the training set and an accuracy of 0.954 on the validation set. This indicates that the model is overfitting the data. On the test set the model achieved an accuracy of 0.931.
Five images containing German traffic signs were downloaded from the web. These images were cropped and resized to 32x32 pixels and then loaded using OpenCV and normalised.
The model achieved 100% validation accuracy. The predictions were as follows.
Image 0: Children crossing
Image 1: Slippery road
Image 2: Go straight or left
Image 3: Priority road
Image 4: Right-of-way at the next intersection
Validation Accuracy = 1.000
Finally the top five softmax probabilities were printed.
Probabilities
[[ 1.00000000e+00 1.92169747e-09 4.02971683e-22 8.78265604e-23 1.16356194e-29]
[ 1.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00]
[ 1.00000000e+00 1.01680692e-21 9.69769272e-24 2.35175907e-30 7.10347746e-34]
[ 1.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00 0.00000000e+00]
[ 1.00000000e+00 6.48320714e-18 1.85014703e-37 0.00000000e+00 0.00000000e+00]]
Predictions
[[28 29 5 24 1]
[23 0 1 2 3]
[37 39 40 4 35]
[12 0 1 2 3]
[11 30 18 0 1]]
A convoluted neural network was created using a LeNet architecture to successfully classify German traffic signs. To improve the model and reduce overfitting the data could be augmented to increase the number of training samples.