Research On TV Seeker Correlation Filter Tracking Algorithm

The TV seeker is a representative of object tracking algorithm in military application.Study of object tracking algorithm with real-time,robustness and accuracy is very important.With the development of machine learning in recent years,there are many trackers arisen.Kernelized Correlation Filter(KCF)is the most popular one for its high performance and super speed in object tracking filed.The work about study and improvement on this algorithm and applying to TV seeker is very perspective.Firstly,this thesis gave some detail theoretical derivations and proofs of KCF,and implementation procedures have been presented.Secondly,since original correlation tracking algorithm cannot adapt to scale variations of the target well,and the scale-pyramid based methods gain higher performance but sacrifice more speed,this thesis proposed a fast scale-adaptive correlation tracking(FSCT)method by establishing a cascade model of target position and scale with the help of log-polar transformation.The benchmark demonstrate that FSCT achieves similar performance with the scale-pyramid based correlation trackers,and performs more than twice as the latter ones in speed.In order to explore the application of this tracker in TV seeker,this thesis took a implementation of KCF with linear kernel function and raw gray features in DM642 platform.Some optimization measures have been taken specifically.Experiments show that this system can track regular target in real time.