Premium
Structure‐independent motion recovery from a monocular image sequence with low fill fraction
Author(s) -
Keshavan Jishnu,
Humbert J. Sean
Publication year - 2017
Publication title -
international journal of robust and nonlinear control
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.361
H-Index - 106
eISSN - 1099-1239
pISSN - 1049-8923
DOI - 10.1002/rnc.3879
Subject(s) - optical flow , differentiator , bounded function , observer (physics) , filter (signal processing) , motion estimation , feature (linguistics) , artificial intelligence , mathematics , computer vision , motion field , computer science , control theory (sociology) , algorithm , image (mathematics) , mathematical analysis , linguistics , philosophy , physics , control (management) , quantum mechanics
Summary Real‐world image sequences are usually characterized by frequent loss and replacement of tracked feature data, and the key to accurate motion recovery from such image sequences lies in the synthesis of a structure‐independent dynamic filter that enables the replacement of lost point features with newly acquired points in a seamless manner. In this note, a novel nonlinear observer is proposed that relies on filtered estimates of optical flow to accomplish structure‐independent motion recovery from monocular image sequences with a low fill fraction. With a single component of linear velocity assumed known, the proposed scheme relies on the perspective observation of at least five points to yield exponentially convergent estimates of the unknown motion parameters that converge to a uniform, ultimate bound in the presence of model error. The unknown linear and angular velocities are assumed to be generated using an imperfectly known model that incorporates a bounded uncertainty, and optical flow estimation is accomplished using a robust differentiator that is based on the sliding‐mode technique. Numerical results are used to validate and demonstrate superior observer performance compared to a leading alternative design on a real‐world–like image sequence that is characterized by significant measurement noise and high feature turnover rate.