SKT-MOT and DyTracker: A Multiobject Tracking Dataset and a Dynamic Tracker for Speed Skating Video

Figure 7

The Kalman filtering process. The KF contains two parts: state prediction and state update. The prediction step uses a constant-velocity model to predict the prior (prediction) state in frame T based on the posterior result in frame T1. The update step fuses new observation with the matched prediction depending on the Kalman gain K calculated, resulting in the posterior state and covariance matrix at frame T.