Winter Training Project

• Developed a traffic-sign recognition modules using The German Traffic Sign Recognition Benchmark.
• Implemented an LeNet to classify 43 traffic-signs.

• Images are 32 (width) x 32 (height) x 3 (RGB color channels)
• Training set is composed of 34799 images
• Validation set is composed of 4410 images
• Test set is composed of 12630 images  There are 43 classes (e.g. Speed Limit 20km/h, No entry, Bumpy road, etc.)
• Image normalisation

While training the network, the following criteria needs to be kept in mind.

• Cost Function
• Optimizer (Adam, Adamax, RMSprop)
• Number of epochs and Batch Size (128 selected)
• Learning Rate Decrease

• train.py - running the keras model
• test.py - testing the accuracy
• utils.py - preprocessing function for histogram normalistaion