Abstract: A robust, adaptive, appearance model is disclosed that includes both a stable model component, learned over a long time course, and a transient component, learned over a relatively short time course (e.g., a 2-frame motion component and/or an outlier processing component). An on-line EM-algorithm is used to adapt the appearance model parameters over time. An implementation of this approach is developed for an appearance model based on the filter responses from a steerable pyramid. The appearance model is used in a motion-based tracking algorithm to provide robustness in the face of image outliers, such as those caused by occlusions. It is also provides the ability to adapt to natural changes in appearance, such as those due to facial expressions, or variations in 3D pose.

Abstract: A visual motion analysis method that uses multiple layered global motion models to both detect and reliably track an arbitrary number of moving objects appearing in image sequences. Each global model includes a background layer and one or more foreground “polybones”, each foreground polybone including a parametric shape model, an appearance model, and a motion model describing an associated moving object. Each polybone includes an exclusive spatial support region and a probabilistic boundary region, and is assigned an explicit depth ordering. Multiple global models having different numbers of layers, depth orderings, motions, etc., corresponding to detected objects are generated, refined using, for example, an EM algorithm, and then ranked/compared. Initial guesses for the model parameters are drawn from a proposal distribution over the set of potential (likely) models.

Abstract: A visual motion analysis method that uses multiple layered global motion models to both detect and reliably track an arbitrary number of moving objects appearing in image sequences. Each global model includes a background layer and one or more foreground “polybones”, each foreground polybone including a parametric shape model, an appearance model, and a motion model describing an associated moving object. Each polybone includes an exclusive spatial support region and a probabilistic boundary region, and is assigned an explicit depth ordering. Multiple global models having different numbers of layers, depth orderings, motions, etc., corresponding to detected objects are generated, refined using, for example, an EM algorithm, and then ranked/compared. Initial guesses for the model parameters are drawn from a proposal distribution over the set of potential (likely) models.

Abstract: A robust, adaptive, appearance model is disclosed that includes both a stable model component, learned over a long time course, and a transient component, learned over a relatively short time course (e.g., a 2-frame motion component and/or an outlier processing component). An on-line EM-algorithm is used to adapt the appearance model parameters over time. An implementation of this approach is developed for an appearance model based on the filter responses from a steerable pyramid. The appearance model is used in a motion-based tracking algorithm to provide robustness in the face of image outliers, such as those caused by occlusions. It is also provides the ability to adapt to natural changes in appearance, such as those due to facial expressions, or variations in 3D pose.

Abstract: A system tracks and identifies view-based representations of an object through a sequence of images. As the view of the object changes due to its motion or the motion of its recording device, the object is identified by matching an image region containing the object with a set of basis images represented by an eigenspace. The eigenspace is generated from a training set of images which records different views of the object. The system identifies the object in the image region by simultaneously computing a transformation that aligns the image region with the eigenspace, and computing coefficients of a combination of linear eigenvectors that reconstruct the image region. This identification and tracking system operates when views of the object in the image are deformed under some transformation with respect to the eigenspace. Matching between the image region and the eigenspace is performed using a robust regression formulation that uses a coarse to fine strategy with incremental refinement.