Files
Download Full Text (4.6 MB)
Abstract
Computer stereo-vision consists on a system that is able to obtain, estimate, and extract distance information of a scene in space from a set of 2D images. The system can be upgraded one step further by utilizing LWIR (Long Wave Infrared) sensors, in order to potentially place the system in low-visibility scenarios where ordinary color sensors might not be the best option. The data streams via an analog signal and is converted to serial-USB to be processed. The sensors are placed parallel, mounted on a bracket and attached to a robotics platform.
In order to estimate depth, a concept called stereoscopic vision is implemented similarly as humans perceive depth. If two sensors are placed into a scene, Epipolar Geometry can be used to obtain distance by calculating the disparity (distance) between matching feature points. The feature points are matched based on the cost from a SAD (Sum of Absolute Differences) algorithm. This algorithm finds a block of pixels from the reference image (Left) and calculates the SAD based on pixel intensities on the target image (Right). This search is simplified by rectifying the two images. This means that each possible feature point is located along the same pixel row, eliminating the need to search vertically. The block with the smallest cost along each pixel row is considered to be the best match. This disparity is inversely proportional to the actual distance of the object (human) and is then sent to the robot in order to track.
Publication Date
2017
Disciplines
Electrical and Computer Engineering | Engineering
Faculty Advisor/Mentor
Dr. Yuichi Motai
VCU Capstone Design Expo Posters
Rights
© The Author(s)
Date of Submission
May 2018