various classes of the image data
distribution plot for image dataset
image preprocessing techniques used
My model
the image is predicted belonging to the exact class speed limit 30km/h

Inspiration

The inspiration behind this project was to work with image data, to find the distribution of the dataset, and to learn the various techniques used for image preprocessing and training a custom convolutional neural network.

What it does

Trained a Deep Learning Model that can identify between 43 different German Traffic Signs using LeNet architecture.

How I built it

The dataset is available at Bitbucket. The dataset is sclassified into 43 different classes: Speed limit (20km/h) Wild animals crossing Turn Left Ahead End of no passing by vehicles over 3.5 metric tons among many other classes.

These are the libraries I have used for this project: 1.numpy

matplotlib
OpenCV
tensorflow
Keras
Pickle
Random
Pandas
Requests
Pillow

The model is based on the LeNet architecture. LeNet is small and easy to understand — yet large enough to provide interesting results and an excellent first architecture for image data and convolutional neural networks. My plot for accuracy vs epoch is

Challenges I ran into

Splitting out Features and labels.
Preprocessing the images- Grayscale conversion, equalisation and image normalisation, splitting the image dataset and fitting it in the imageGenerator.
working on the model architecture, activation functions, and training the dataset for specified number of epochs

Accomplishments that I'm proud of

I've understood various image processing algorithms, model architecture and that I was able to predict the image oytcomes with an accuracy close to 98%

What I learned

I have learnt ways to work and classify image data, the various CNN architectures and the choice of various hyperparameters for tuning the model

here is the code collab

What's next for Behavioral Cloning for Self Driving Car

This data project was a small part of the self driving car project that I simulated in a simulator. This trained model can be further processed and trained on a self driving car model so that it can be simulated by obeying various traffic rules like speed limit etc.