Automated switchover between video tracking modes with Kalman filtering and particle filtering for optimal performance

An approach is developed for ensuring the optimal tracking performance by separately taking algorithmic strength identified with Kalman filter (KF) and particle filter (PF). A parameter is engineered to on-line evaluate the performance of KF. The threshold is derived for the parameter to activate the switchover between KF and PF. The proposed approach is applied in simulation with a given dynamic model and in a real video sequence to track a target ball, in which the required conditions for KF to perform are purposefully violated. The experiment results suggest that the KF tracking mode can be automatically switched to the PF tracking mode when the required conditions are not met for KF, suggesting the feasibility of the presented method.