Based On Multi-level Feature Series Correlation Particle Filter For Visual Target Tracking

Visual target tracking technology mainly include image processing,pattern recognition,deep learning and artificial intelligence.It belongs to a vital research content in the current computer vision field.Because of overcoming targets change and random tracking environment,scholars are committed to researching a visual target tracking algorithm with superior performance.This paper is devoted to improving the robustness and accuracy of the visual tracking algorithm,based on the improved particle filter algorithm and correlation particle filter algorithm of multi-level feature series:1.In view of the great limitations of traditional single-feature visual target tracking algorithms,proposes a new target tracking algorithm framework: a multi-level feature series algorithm framework.In the first filtering,the CN feature is selected for the target extraction.According to this feature,in the first filtering the target is predicted and estimated,and the updated position of the target is determined;the second filtering is based on the first filtering,and the edge feature is selected for the target to extract.The feature predicts and estimates the target again to obtain a more accurate updated position of the target,thereby determining the predicted position of the target in the next frame.2.Because the traditional particle filter algorithm is prone to reduce the problem of particle diversity,proposes a method to improve its resampling step.After the importance sampling is completed,a set of particle sets with different weights are obtained.From these particle sets,the particles with larger and smaller weights are selected,and new particles with larger weights are randomly generated in the surrounding area.And the particles with smaller weights are eliminated,the final number of particles remains the same as before.And combined with the multi-level feature series framework,which is the multi-level feature series particle filter algorithm.This algorithm not only effectively overcomes the loss of particle diversity caused by the traditional particle filter resampling step,but also greatly improves the efficiency of particle filter calculations.Experiments show that this improved algorithm can solve the problems of target deformation,occlusion and background clutter.3.Due to the shortcomings in the application of traditional particle filtering and correlation filtering algorithms,proposes a multi-level feature series correlation particle filter algorithm.The particle filter algorithm can deal with target scale changes and occlusion issues.The correlation filter algorithm can sample the particles are guided to the target state for distribution mode and training features,reducing the amount of calculation.Experiments show that compared with other algorithms,the multi-level feature series correlation particle algorithm can improve the accuracy and robustness of the tracking task.This paper not only guarantee accuracy under the influence of many challenging factors,but also reduce the complexity of the particle filter algorithm.Compared with other algorithms,this paper have superior accuracy and robustness of the tracking task.