A method for tracking the pose of known 3-D objects based on an active contour model

A method for tracking the pose of known 3{D objects moving with 6 degrees of freedom in space is described. The main idea is to split the tracking into two quasi autonomous working processes, one to track the 2{D outline (i.e. silhouette) of the object in an image sequence and one to track the full 3{D object pose based on the estimated 2{D outline. An active contour model utilizing a Kalman lter is used to eeciently track the object outline represented as contrast edges in the image sequence. Geometric information used are limited to the space curve at the object's surface projecting to the currently visible outline. The 3{D pose tracker uses the results of the contour tracker, i.e. the position and shape of the outline, and an internal 3{D object model to track the full object pose. The object pose is derived from an n{point correspondence between the outline and the object's surface. Furthermore, the pose tracker predicts the object appearance, i.e. the object aspect, and provides the contour tracker with new model information on object aspect changes. The presented results show the successful application of the method to tracking of polyhedral objects in image sequences.