Why I'm Researching Urban Navigation with Street View Imagery
A short introduction to my M.Sc. thesis at BUET: can models learn to find their way through a city using only street-level images?
By Dewan Tanjil Hossain ·
Humans navigate cities by recognizing landmarks, reading signs and remembering what a street looked like. Can a model do the same from street-level imagery alone?
The question
My thesis builds a benchmark using Google Street View imagery to evaluate how well models navigate urban environments over space and time.
What makes it hard
- Streets look different across seasons and years
- Many intersections look alike
- Long routes require memory, not just perception
Where it stands
The project is currently in the data collection phase. I'll share updates on the dataset design and early baselines here.