Detection of independent motion using directional motion estimation

It is shown that the problem of independent motion detection can be address ed by analyzing constraints on low-dimensional directional (projected) comp onents of flow fields. We construct a robust algorithm, implemented as a re cursive filter, to extract directional motion parameters from long image se quences. Based on this, a qualitative approach is described to detect indep endent motion, involving a combination of robust line-fitting and one-dimen sional search. The low-dimensional projections onto subspaces facilitate ef ficient dynamic self-adaptation of detection thresholds to achieve good per formance under changing operational conditions. The analysis is extended to long image sequences by incorporating tracking and spatio-temporal filteri ng. The approach is applicable to general camera motion and cluttered scene s using a wide range of camera fields of view. We demonstrate it on a varie ty of real image sequences. (C) 1999 Academic Press.