Method and apparatus for tracking moving objects in a sequence of two-dimensional images using a dynamic layered representation

摘要

A system that tracks one or more moving objects in a sequence of video images employs a dynamic layer representation to represent the objects that are being tracked. The system concurrently estimates three components of the dynamic layer representation—layer segmentation, motion, and appearance—over time in a maximum a posteriori (MAP) framework. In order to enforce a global shape constraint and to maintain the layer segmentation over time, the subject invention imposes a prior constraint on parametric segmentation. In addition, the system uses a generalized Expectation-Maximization (EM) algorithm to compute an optimal solution. The system uses an object state that consists of representations of motion, appearance and ownership masks. The system applies a constant appearance model across multiple images in the video stream.