Once some potential targets have been detected in each image, they need to be linked together to describe the path that the detected object took, which can then be analyzed to help understand the behavior of the target. We assume an relatively sparse scene with no occlusions.