3D mapping and depth mapping of environments using stereo cameras.
The framework used first obtains depth and disparity maps through algorithms like Census Transform(CT) as well as Sum of Absolute Difference(SAD) Algorithm to implement block matching. These disparity maps are then converted into point clouds using Open3D. The results obtained from the different methods are also observed, compared, and the results are shown. These algorithms are first tested on stereo datasets before being implemented in the Gazebo simulation environment. A novel method is also proposed which reduces the errors when compared with the existing algorithms. In contrast to common existing methods used to calculate disparity maps, we have used Multi-Block Matching along with SAD and CT to perform block matching. This reduces the bad error when compared with the ground truth.
This project was done for Image and Video Processing Course from Jan-May 2021.
Project Members:
Roshan Rangarajan
Saikumar Dande
Chandravaran Kunjeti