Model-based tracking of 3D objects based on a sequential Monte-Carlo method

One details here a method to track a 3D object and estimate its 3D position and motion parameters from a monocular image sequence. This estimation problem is modeled by state equations that describe the dynamics of the object and the measurement delivered by the sensor. One develops a top-down approach that needs a point description of the shape to track. This allows a direct comparison with the pixels in the image without any preprocessing that may give rise to additional errors. The proposed method also delivers an estimation of the dense 3D motion vector field and, by projection onto the image plane, of the 2D motion field that can be compared with the optical flow methods.