Research On The Method Of Target Tracking Under The Conditions Of Occlusion And Scale Variation

Visual target tracking technology is a research hotspot in computer vision,which has a wide range of applications in the fields of intelligent monitoring,human-computer interaction,intelligent transportation and drones.With the development of computer technology,researchers have proposed many excellent theories and tracking algorithms to improve the accuracy and robustness of tracking.However,in the actual tracking scenes,due to the influences of factors such as occlusion and scale variation,tracking drift will often be caused and ultimately lead to low tracking accuracy or even tracking failure.Based on particle filter and kernel correlation filter tracking algorithm,this thesis studies the improvement of anti-occlusion particle filter algorithm,scale adaptive estimation of block kernel correlation filter and template updating optimization.The details are as follows:1.The particle filter robust tracking method under occlusion is studied.Aiming at the poor target tracking accuracy problem caused by particle impoverishment in particle filter under the occlusion condition,an anti-occlusion chicken swarm optimization-based particle filter tracking method is proposed.Firstly,the chicken swarm optimization algorithm is integrated into the sampling stage of the particles,that is,the weight of particles is used as the fitness and the type of each particle in the population and interrelation among particles is determined,and various movement mechanism of different types of particles are introduced to update the position.Then individual learning is adjusted through linear decreasing weight and the global optimal learning strategies to solve the local optimal problem,and we select the area which has the biggest likelihood function value as the target location.Finally,the occlusion tracking is continued by the template updating.2.The adaptive scale tracking framework of sub-block joint estimation based on kernel correlation filter is explored.In order to solve the problem that the anti-interference ability of kernel correlation filter tracking algorithm is not strong and it is easy to be interfered by external factors such as occlusion and scale variation,the tracking accuracy is decreased.A scale-adaptive kernel correlation filter tracking method with improved block strategy is proposed.Firstly,the target is adaptively divided into two blocks according to the geometric characteristics of the tracking frame,and the corresponding response map is obtained by independently tracking on each sub-block using kernel correlation filter,and then the target position is estimated by calculating the weight and the deformation vector of the sub-block relatively to the target.Finally,the realization on the overall optimal candidate target scale is estimated by using the scale factor.3.The method of template updating optimization in the kernel correlation filter framework is researched.The template drift problem caused by the occlusion noise information is introduced into the template updating process of the kernel correlation filter algorithm in the occlusion and scale variation scenes.Based on the scale-adaptive kernel correlation filter tracking method with improved block strategy,an occlusion detection method for jointing the peak value of correlation filter response and average peak-to correlation energy determination is proposed.Firstly,the two index values of the peak value of correlation filter response and the average peak-to correlation energy of the sub-block are calculated,and then the corresponding threshold judgment result is used as the criterion for template update to realize the adaptive update of the template.