ind1010 / SfM_for_BusEdge

Code and data for the paper "Leveraging Structure from Motion to Localize Inaccessible Bus Stops" for RISS 2022

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Leveraging Structure from Motion to Localize Inaccessible Bus Stops

Authors: Indu Panigrahi, Tom Bu, and Christoph Mertz
Paper available here

Contents:

  • BusEdge/ - This directory contains scripts for obtaining and filtering images from the BusEdge Platform.
  • Mask2Former/ - This directory contains the segmentation model that we use (Mask2Former) and scripts for detecting snow coverage. See Mask2Former/README.md for more information.
  • colmap/ - This directory contains sample images (references + query) and the associated reconstruction in the form of text files. These are used by the scripts in Mask2Former/.
  • other_files/ - This directory contains files that list some information about image sequences and calendar dates with snowy weather in Pittsburgh.

Instructions:

Here is a dataset of the images used in this work and along with images in additional categories

  1. Run COLMAP on clear weather images from a scene and then re-run COLMAP with the query added. Save the resulting reconstruction as text files using the naming conventions that we use in the sample files and folders that we provide in colmap/.
  2. Run Mask2Former/estimate_ground_plane.ipynb. We provide sample output from this notebook in Mask2Former/sample_output/.
  3. Separate the query image from the clear weather images into the directory structure demonstrated in Mask2Former/images/.
  4. Run Mask2Former/detect_coverage.ipynb. This notebook performs the detection of snow-covered sidewalks.

Requirements:

  • Used CUDA 11.1
  • See requirements.txt

Reference:

Please cite this code as follows:

@article{panigrahi2022riss,
  author  = {Indu Panigrahi and Tom Bu and Christoph Mertz},
  title   = {{Leveraging Structure from Motion to Localize Inaccessible Bus Stops}},
  journal = {{Robotics Institute Summer Scholars Working Papers Journal}},
  year    = 2022,
  volume  = {10}
}


@mastersthesis{bu2022crosswalk,
author = {Tom Bu},
title = {{Towards HD Map Updates with Crosswalk Change Detection from Vehicle-Mounted Cameras}},
year = {2022},
month = {August},
school = {Carnegie Mellon University},
address = {Pittsburgh, PA},
number = {CMU-RI-TR-22-34},
}

About

Code and data for the paper "Leveraging Structure from Motion to Localize Inaccessible Bus Stops" for RISS 2022


Languages

Language:Jupyter Notebook 75.7%Language:Python 14.0%Language:JavaScript 7.7%Language:Cuda 1.2%Language:C++ 0.5%Language:CSS 0.3%Language:Dockerfile 0.3%Language:Shell 0.2%Language:HTML 0.1%Language:CMake 0.1%