3.3 Recursive Tracking
A simple strategy can be used for tracking shapes on a frame to
frame basis, which allows skipping the recognition step for most
shapes in most frames. This is achieved by maintaining for each
shape a set of simple properties that can be efficiently and
robustly tested to see if the shape corresponds to any shape in the
previous frame. For each image shape, we calculate its centroid,
length, and area. In each frame we attempt to find the
corresponding shape from the previous constant number of frames
according to these properties.
In fact, we can skip the DLT step as well, using the
transformation calculated for