In this research project, I investigate the possibilities of making a car make and
model recognition mobile application, by the use of Machine Learning as a Service.
Focusing on tools provided by IBM, Amazon, Microsoft and Google, I will test and
analyse each of these before making a decision on which solutions to proceed with
during the implementation and development of a mobile application.
During this thesis project I analyse some of the important histories and break-
throughs within AI and Machine Learning, and what challenges there has been and
still are. CNN and Deep Learning are very hot topics and many MLaaS services are
still in beta or in the early stages.
Testing the dierent solutions created, I decided to use Microsoft CNTK for Image
classication along with Google OCR for text recognition. Using an open source
library like CNTK allows for more user specications which led to an overall image
classication of around 93% and being a Microsoft Azure product, there was a lot of
documentation on getting it published as a RESTful API. Testing Google OCR for
license plate character extraction resulted in a clear winner. It extracted all characters
and sometimes more, which I process further in the application development section
before making HTTP requests to my license plate data provider.
I do believe there is potential for using MLaaS services for CMM surveillance.
More conguration needs to be done as well as further testing. I suggest a much
larger data set with more specic features such as cropped images with not that
much background noise as well as more specic images of logos and head- and tail
lights shapes. Using grayscale or black/white images might also aect the results
since this could give less interference between the car models.
Keywords: ANN (Articial Neural network), CNN (Convolutional Neural Network), Android
Mobile Application, MLaaS (Machine Learning as a Service, Automatic License Plate Recognition,
Car classication, Optical Character Recognition
|Uddannelser||Datalogi, (Bachelor/kandidatuddannelse) Kandidat|