Team Members: Alper Adil Çetinkaya, Esme Şule Alp
Image based localization is one of the popular and succesful way of using machine learning technics. There are many publications, different attempts and developing results in this area. Simply, it takes image of a place and predicts where it is.
Different from previous works, we’re going to make image based localization in the streets of our home town, Ankara. To do this, we will need a large dataset of different places in Ankara; because we have to teach the machine first and the more data we have, the better it will learn. We haven’t decided how to compose our dataset yet. It may be walking the streets and taking pictures or it may be using Google maps.
One thing that we wish to do also classify if the image contains a single prominant building or not. If so, than we wish to find the address of that building and present it. If the building is a well known structure or a workplace than that should also be spesified. If there is no single prominant structure in the picture than the general location of possibly just the street name or province will be given.
This week we discussed what we can do to differentiate our project from those that have a similar topic, we weighted in what is possible and what is probable and decided to pursue on a single city. We do not know if we can actually pull off determining just any building in the city; but researching that to see if it’s plausable, along with other research that needs to be done is left for another week.
