Doshi, A.Starck, J.Hilton, A.2025-12-172025-12-172010Journal of Virtual Reality and Broadcasting, 2010; 7(3):1-201860-2037https://hdl.handle.net/1959.8/122837This paper presents an empirical study of affine invariantfeature detectors to perform matching on video sequencesof people with non-rigid surface deformation.Recent advances in feature detection and wide baselinematching have focused on static scenes. Videoframes of human movement capture highly non-rigiddeformation such as loose hair, cloth creases, skinstretching and free flowing clothing. This study evaluatesthe performance of six widely used feature detectorsfor sparse temporal correspondence on singleview and multiple view video sequences. Quantitativeevaluation is performed of both the number of featuresdetected and their temporal matching against andwithout ground truth correspondence. Recall-accuracyanalysis of feature matching is reported for temporalcorrespondence on single view and multiple view sequencesof people with variation in clothing and movement.This analysis identifies that existing feature detectionand matching algorithms are unreliable for fast movement with common clothing.enFeature matchingsiftvideo sequencesJournal article