*Project developed while a grad student at NYU ITP

3D model of a section of 183rd street in Queens, New York

Inspired by the NY Times work using 3D models, I proposed Documented the use of a 3D model in their climate change reporting, especifically on a story about a street in New York City that consistently experiences flooding. Unfortunatelly, due to web developping issues, the 3D model couldn’t be published when the story ran in April 2023. However, the process of creating and optimizing a 3D model taught me many technical lessons that I plan to use in future projects.

I used photogrammetry, a technique that uses still images to create a 3D model. I scanned 183rd street, between 90th and 91st avenues, in Jamaica, Queens, New York. This section of 183rd street is about 270ft long with two story homes on both sides.

I used three DSLR cameras on a rig similar to this one (NY Times R&D). I used a similar rig on previous (though smaller) photogrammetry projects and I found it to be very helpful.

I captured a total of 5272 images in raw format, which were then processed in Photoshop before sending them to Reality Capture for reconstruction.

I shot more than 5000 images using three canon DSLRs each with a 16-35mm lens

Original Reality Capture reconstruction - 656 million tris and 680 8K textures

This video shows the final optimized model. The original plan was to publish this model with an overlay of text and photographs to explain the consequences of flooding in this street.

Lessons learned

The entire process was very time-consuming. It takes time to process 5000+ raw mages and convert them to jpg. Additionally, I had to add control points in some areas where Reality Capture didn’t do a good job with the alignment. But, by far, the most complex and time-consuming part of the process was the optimization and simplification. The characteristics of the street didn’t help. The street had cars, fences, gates, lightpoles - elements that add significant complexity to the scene. Considering that this model was to be embedded on the web, optimization was critical. Advanced knowledge of Maya or Blender is necessary.

I chose a cloudy day to take the photos, which was very helpful considering that the entire capture took about three hours. Once I started taking photos, some of the parked cars in the scene were moved. Everytime this happened (about 4 times) I took note and went back to that area to take more photos of the “new” scene. Also, when I processed the photos I had to edit out some photos that included cars that were moved. The notes I took on the field were extremely helpful.