Create a unsupervised CNN for traffic sign recognition

For my undergrad program I have to create a CNN for traffic sign recognition in the Netherlands. I have picture data from a moving car with camera's mounted on it that has taken 100.000+ pictures. This data is unlabeled so unsupervised learning has to be used. An example of pictures is given below in a document. There is another data set that is available online that has closeup images of dutch traffic with labels, but these have to be extracted from URL links from an excel file.

Due to some issues the project has to be done in 3 days.

The outcome should be an algorithm that can recognize the type of different traffic signs in The Netherlands.