User Tools

Site Tools


projects:year10:10a.014.sbu

10a.014.SBU - Object Recognition Under Various Illumination Conditions

Project - Summary

The goal of this multi-year project is to design a convolutional neural network (CNN) using Tensorflow and/or Caffee to perform item recognition under various illumination conditions. The challenge is to perform item-recognition (inference) + other to be determined techniques in near real-time for partially obstructed items typically seen in realworld applications across various lighting conditions. Definition of the use case, examples of obstructed images (items held in hands or covered by clothing), camera hardware, and compute power will be provided by the project sponsor.

This is a multi-year project started in Year 9.

Project - Team

Team Member Role Email Phone Number Academic Site/IAB
Dimitris Samaras PI samaras@cs.stonybrook.edu (631) 632-8464 Stony Brook University
Arie Kaufman Co-PI ari@cs.stonybrook.edu (631) 632-8441 Stony Brook University
Shawn Mathew Student shawmathew@cs.stonybrook.ed (516) 306-1876 Stony Brook University
Georgi Georgiev Student ggeorgi@cs.stonybrook.edu N/A Stony Brook University
Jingyi Xu Student jingyixu@cs.stonybrook.edu N/A Stony Brook University
Miroslav Trajkovic Project Mentor Miroslav.Trajkovic@zebra.com (631) 245-5352 Zebra
Funded by: Zebra

Project - Deliverables

Deliverables
1 Dataset Creation
2 3D Reconstruction
3 Design & validate object recognizer baseline
4 Shadow & shading removal model
5 Combine illumination correction & object recognizer

Project - Novelty of Approach

Stony Brook project will provide convolutional neural network (CNN) framework to be used to recognize items in near real-time for partially obstructed items typically seen in real-world applications across various lighting conditions, while current state of the art employs computer vision techniques.

Project - Benefits to IAB

  • Object recognition model with application in real-scenarios where non-uniform illumination effects its performance.
  • Shadow, shading, and illuminant removal framework will be useful to pre-process images for various other tasks.
  • Proposed dataset will be helpful for not only object recognition but also to learn different tasks such as shadow removal, object segmentation, etc.

Project - Documents

This is a multi-year project started in Year 9.

Below is a project update presented at the 2021 CVDI IAB Spring Meeting.

10a.014.sbu_multiple_year_project_slides.pptx
10a.014.sbu_zebra_cvdi_year_10_final_project_report.pdf

projects/year10/10a.014.sbu.txt · Last modified: 2022/10/03 15:24 by sally.johnson