Exploring Visual Motion Using Projections of Motion Fields

The dimensionality of visual motion analysis can be reduced by analyzing projections of ow vector elds. In contrast to motion vector elds, these projections exhibit simple geometric properties which are invariant to the scene structure and depend only on the camera motion. Using these properties , structure and motion can be either completely or partially decoupled. We estimate motion parameters from projections of ow elds by using robust techniques, implemented in a recursive observer model. The recovered scene structure is collected in a visual scene memory for both temporal and spatial integration of instantaneous measurements. The model is applicable to general camera motion and to almost arbitrary known focal length without the requirement of point correspondence. We demonstrate our projection method on the problem of detecting independently moving objects from a moving camera. Using the projection approach, the problem can be reduced to a one-dimensional optimization process which involves robust line-tting and outlier detection.