Facial Feature Reconstruction using Structure from Motion

[摘要] Structure from Motion suggests that an object or scene's three-dimensionalstructure can be determined from its observed two-dimensional motion. Humanefforts, manifested in computer algorithms, try to mimic the enormouspower of the visual processing capabilities of the human brain. We presentan algorithm to estimate structure, using the Unscented Kalman Filter, fromthe motion of point-wise features, produced by the Kanade-Lucas-Tomasi featuretracker. The algorithm is evaluated critically against an extensive set ofmotion sequences, with special attention paid to facial feature reconstruction.

[发布机构] Stellenbosch University