Discipline: Technology and Engineering
Subcategory: Computer Engineering
Session: 3
Jatwon Antonio Burkes - Bethune-Cookman University
Co-Author(s): Xiaohe Wu, Bethune-Cookman University, Daytona Beach,FL; Juan Calderon, Bethune-Cookman University, Daytona Beach, FL.
It is possible to construct a 3-D model of the world from 2-D visual information.
In the last decades, the advances in the use of GPS have helped robots navigate autonomously in open areas. However, this type of system presents problems when it is not possible to have access to GPS information or when the area where the robot plans to navigate is indoor or more accurate and richer information is needed. A stereo perception is a great tool for robotic localization and navigation system, for this reason, we propose the generation of a perception system based on stereo vision.
In this project, we used 2-D visual data and extracted the depth information using the Semi-Global Block Matching (SGBM) algorithm and the result is displayed as a disparity map. Before doing the 3-D reconstruction, the cameras that were being used were calibrated to compensate for optical distortion. A mathematical model of radial and tangential distortion is established; also, the computer code was developed to find the parameters in the model. A stereo vision system was set up to capture a 2-D image from the camera on the left-hand side and a 2-D image from the right-hand side. Then, the SGBM algorithm was applied to find out the depth information of the 2-D images.
The experiment’s results indicate that the algorithm work fine in general to give the depth information in the real world. However, the result is highly sensitive to noise especially when flat surfaces with uniform texture are place in front of the camera.
For the future work, we plan to improve the algorithm by employing better feature extraction methods, such as deep neural network.
Funder Acknowledgement(s): This research project was supported by the National Science Foundation Research Experiences for Undergraduates (REU) Award Number DMS-1755561 ; Bethune-Cookman University, and Dr. Seenith Sivasundaram and Dr. Yungkul Kim for facilitating the REU Program
Faculty Advisor: Xiaohe Wu, wux@cookman.edu
Role: The part of the research that I have done was the literary reading and programming (in Python). I was required to read a lot about visual data. Soon after reading, we noticed that stereo visual is good for reconstructing 3D images. The part I spend the most time on was programming and testing. I also found a library called OpenCV, which helped a lot with visual data. Once I have read about that and looked further into it, I asked Dr. Wu about his knowledge of this library and he gave me great advice, along with Dr. Calderon.