A method, apparatus, and computer program product for joint kinematic and feature tracking are presented. Kinematic measurements and feature/class measurements regarding an object are received from a sensor. A probabilistic argumentation operation is performed on the feature/class measurements using information from a knowledge base and a track file ect to generate feature track likelihood scores regarding likely tracks for the object. Kinematic track likelihood scores are generated based on the kinematic measurements and the track file. Joint track likelihood scores are generated based on the feature track likelihood scores and the kinematic track likelihood scores. Joint track likelihood scores are used to generate a multi-frame track measurement association to determine a most likely track for the object. The track file is continually updated with the most likely track for the object, so that the most likely trajectory of the object is obtained.