• car
• vehicle
• license
• damage
• classification
• detection
• model
• pov
• plate
• image

In the last years the use of machine learning algorithms is grown at a very high
speed, involving a lot of different fields and applications. In this theesis the main
goal is to develop a solution which is able to detect and classify some features
of a car, starting from an image. In particular is required that the solution is
able to fullfy this requiesites:
• Starting from a generic image, detect the portions related to a generic car.
• For each portion, classify the model of that car and the point of view of
the image.
• Detect and obsure the license plate.
• Detect the exterior damages.
The process followed to reach these results started to the analysis of similar
works in the literature, studying some papers in search to a possible approach
(the main works studied are presented in a specific chapter ), passing then to
the choice of the more appropriate approach, and adapting it to fullfy all the
previous requisites.
Before to develop the solution, it was also analyzed what are
the main tools that fit our needs, in order to realize a solution using innovatives
platfoms and approaches.
The solution adopted is based on a classical convolutional neural network to classify the simplest
element of the image, and a Faster R-CNN to detect license plates and damages.
The final part of this work is the analysis of the obtained results, in order to be sure that the project presented is working fine.