The system performs Target Tracking Video Stabilization, where we first define the target to track and we establish a dynamic search region, whose position is determined by the last known target location. We then search for the target only within this search region, which reduces the number of computations required to find the target. In each subsequent video frame, we determine how much the target has moved relative to the previous frame. We use this information to remove unwanted translational camera motions and generate a stabilized video.

Groups of connected pixels are detected using blob analysis, which are likely to correspond to moving objects. The association of detections to the same object is based solely on motion, estimated by a Kalman filter, an optimal method for tracking linear dynamical models under the assumption of Gaussian noise. The filter is used to predict the track's location in each frame, and determine the likelihood of each detection being assigned to each track.