It has been more than a year since two powerful earthquakes struck the southeast of Turkey, taking more than 60,000 lives overnight and causing $34.2 billion in damage. Despite being some of the largest earthquakes recorded in the region’s history, multiple reports attribute the scale of damage to widespread construction malpractices. These malpractices, orchestrated by corrupt officials in the construction industry, result in a common set of structural defects that are visible even to the bare eye. More pressingly, seismologists predict there’s a 70% chance that an equally powerful earthquake will hit Istanbul in the next five years, a city where 20 million people live in 1.2 million buildings, more than half of which are expected to sustain significant damage — if not collapse completely. At the current rate of municipal structural testing — 120 buildings per day — it may take up to 21 years for a resident to see if their building is safe.
This is the main underlying challenge that lead to this experiment, with the aim of combining remote sensing practices with recent multi-modal segmentation algorithms, to look for structural deficiencies across millions of publicly-accessible street imagery.
Ultimately, this project aims to further develop an AI-powered novel geaospatial workflow to get useful insights about the built environment, generating large-scale datasets across regions where street imagery data is available.
Developed by Taha Erdem Ozturk at the Center for Spatial Research, Columbia University.