A Study On Object Tracking Methods Based On Swarm Optimization Algorithm

Visual tracking is a hot topic in computer vision,as the development of the computer,the visual surveillance network and machine learning.It has a wide range of applications in military and urban fields.Although some visual tracking methods have been proposed over the last 20 years,there are still lots of the challenging problem to be solved such as the severe partial occlusion,drastic illumination changes,abrupt motion and motion blur et al.,so it is an important issue how to design a robust and general visual tracking algorithm.1.A tracking method based on ant lion optimization(ALO)algorithm has been proposed.Ant lion algorithm was introduced into the visual tracking framework as a search strategy.And the parameters adaptive and sensitivity of the tracking method were studied.Experiments show that the proposed tracking method had better tracking results,especially when the object has an arbitrary motion or undergoes abrupt motion.2.A tracking framework based on Extended Cuckoo Search(ECS)and Kernel Correlation Filtering(KCF)has been proposed.In this framework,CS was introduced into the KCF tracking framework,and the switching model was set up to form a tracking framework that can track both smooth and abrupt motion.In addition,the poor convergence ability of the CS lead to accuracy was reduced.The simplex method with powerful local optimization ability was used to improve CS,which enhanced the convergence ability of the algorithm and improved the tracking accuracy.3.The performance of three tracking methods based on classical group optimization algorithm and a tracking method based on single solution are compared.The tracking experiments based on ALO,CS and Particle Swarm Optimization(PSO)algorithm were carried out,and their tracking results were compared with those based on Simulated Annealing(SA).And the advantages and disadvantages of each method in the specific tracking scene are explained,which provides the possibility to design the tracking method of the specific scene in the future.