This is a simple project which implements a Traffic Sign Classifier using a Convolutional Neural Network - LeNet5 Architecture.
The purpose of the project was to get familiar with Keras.
Sample data gathered from The German Traffic Sign Recognition Benchmark.
The network is made up of 6 hidden layers, one input layer and one output layer.
Layer | Type | Size | Filter Size | Filter Stride | Padding |
---|---|---|---|---|---|
Input | - | 32x32x3 | - | - | - |
Hidden 1 | Convolutional | 28x28x6 | 5x5 | 1 | Valid |
Hidden 2 | Max Pooling | 14x14x6 | 2x2 | 2 | Valid |
Hidden 3 | Convolutional | 10x10x16 | 5x5 | 1 | Valid |
Hidden 4 | Max Pooling | 5x5x6 | 2x2 | 2 | Valid |
Hidden 5 | Fully Connected | 120 | - | - | - |
Hidden 6 | Fully Connected | 84 | - | - | - |
Output | Fully Connected | 43 | - | - | - |
Model Accuracy: +90%