Our tool of choice was MATLab so all code is written in MATLab language.
Our goal was to design a neural network which can identify different german speedsigns. First we build up a data collection where we collected pictures of different signs. Furthermore we classified the data in "30 km/h","50 km/h","60 km/h" and "No speedsigns/other signs". We constructed a deep network in different variants. First we used a faster RCNN (ResNet50) for region detection linked with a simple CNN for classifying data. Maximizing acurracy was our goal while we optimized parameters.
Second try was a setup with a faster RCNN (ResNet50) for region detection linked with a AlexNet for classifying data. The AlexNet is a neural network which performs very good in recognizing objects. We used a technique called transfer learning in which we relearned the last three layers. With this we optimized accuracy of our deep learning network significantly.