3D digital model reconstruction of insects from a single pair of stereoscopic images

Computer Graphics;Computer Vision;Digitization and Image Capture;Three-Dimensional Graphics and Realism;Stereo Vision

Lau, JohnMark ; Hart ; John C.

关键词: Computer Graphics;    Computer Vision;    Digitization and Image Capture;    Three-Dimensional Graphics and Realism;    Stereo Vision;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/46786/JohnMark_Lau.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

This thesis proposes a robust method of creating a 3D digital model of an object through stereoscopic reconstruction. There are existing methods and tools for achieving this, but they tend to only perform well under favorable conditions, such as specializing to a specific class of objects or requiring certain geometric and textural characteristics. Hence, when the object in question has a relatively monochrome color, minute distinctive features, curved or unusual shapes, or specular surfaces, the results are often warped and do not resemble the original object. A prime example of this loss of digital fidelity can be found in the insect kingdom. Insects come in an enormous variety of shapes and sizes, providing a sizeable database that can rigorously test the algorithm. Our method takes a pair of stereo images and a set of calibration images, finds dense correspondences between the two images, and outputs a colored 3D point cloud. The results are a remarkable improvement over existing tools that attempt to create models for a generic object. The generated models capture features from the original insects and show significant detail given the challenge of the generalized problem.