The high-level software architecture is depicted on figure A.1 on page .
The basic idea is that the program will get an input image (test car). The program has a database of car images which are used for training. The car make is known for these vehicles to the program. We expect the program to display a car make for the test image.
During the classification process there is an feature vector cache created from every train car. This helps to speed-up the overall classification process.