Rutgers RGB-D Amazon Picking Challenge
Zach, 01 April 2020
This was a class assignment where we had to perform pose estimation on an object that was store din a shelving unit using only RGB-D data. The assignment was based on the Amazon Picking Challenge and featured data collected by Rutgers. The pose estimation for 6 objects was determined using the SIFT algorithm which determined similar features from both the object face and the object located in the bin then the homography matrix was found using RANSAC. This project was done entirely with the Python implementation of OpenCV.
The results we have can be seen in the formal report found in my github here. The algorihtm worked for some objects but objects with similar faces on every side and small objects proved to be difficult to determine pose. In addition, non-rectangular object proved the most difficult.