Ανάπτυξη συστήματος αυτόματης αναγνώρισης graffiti χρησιμοποιώντας αλγόριθμους βαθιάς μάθησης

Abstract

In the scope of this thesis, Machine Learning technologies are studied upon, as well together with hardware and software frameworks that support each technology. The practical goal is to make an automated graffiti tag detection application, with which, the user can use a smartphone to detect graffiti tags in real-time through the use of the device’s camera. The development of such application involves effort in creating a large dataset containing images with graffiti tags taken in different environments, as well as annotating each image with bounding boxes indicating the location and size of each tag. The dataset is adapted to the requirements of the software framework used. Various machine learning techniques are used, such as transfer learning to minimize training time and improve accuracy, of the detection system in order to attempt to get the best experience possible from it.