Research On Particle Filter Target Tracking Algorithm Based On Multi-feature Fusion

Computer vision includes many fields,moving object tracking has become a popular research direction and hot spot.Range of use and application area are extremely a lot,such as intelligent monitoring,human-computer interaction,robot navigation,flow control,bio-medical diagnosis,etc.So far,although domestic and foreign experts and scholars have put forward a lot of classical algorithms and improved algorithms based on them,but these proposed target tracking algorithms are still faced with enormous challenges.For instance,illumination changes,occlusion,nonlinear deformation caused by the sudden movement of the target,background and target tracking have high similarity and noise interference in complex background,etc,which greatly affect the speed and accuracy of these target tracking algorithms.It may cause incalculable consequences.Considering these problems,it is difficult and arduous to realize a target tracking algorithm which can adapt to complex scenes.At present,Particle filter is the best method to solve non Gauss nonlinear tracking problem.The algorithm can well adapt to a variety of external interference factors and can ensure the accuracy and robustness of target tracking.(1)But the classical particle filter target tracking algorithm uses a single feature to describe the target.which is easy to be affected by light,occlusion,the similarity of object and background in the tracking process.In order to solve this problem,a variety of target features are fused to particle filter for target tracking.In the framework of particle filter,comprehensively consider the robustness and accuracy of various characteristics to different interference factors to select the appropriate feature extraction algorithm to get into it.Using the dynamic calculation method to calculate the discrimination and stability of adopting difference feature extracting algorithm on between object and background,automatic select the feature of the high discrimination,good stability to characterize the target,and then form a multi-feature fusion target model.the extracted feature weight is dynamically adjusted using the uncertainty of the measurement.Dealing with the change of the deformation and the scale of the target by using the color feature of the particle filter algorithm,the edge feature information to adapt changes in the background and the Local Binary Patterns(LBP)features and the image of the gray level information to adapt illumination changes.Which makes the improved algorithm has better stability and accuracy.(2)Although the above algorithm can well adapt the change of most of the external environmental factors.But when the target is occluded in video sequence,tracking excursion or tracking loss is prone.Therefore,using the contextual information of tracking target to solve the problem of target occlusion in the framework of particle filter.Because the relationship between the object to be tracked and the context of the local scene is similar.Therefore,the position of the target in the next frame can be estimated by analyzing the context information of the target in the previous frame.Fast and robust tracking is achieved when the target is occluded.Finally,the tracking performance of the proposed target tracking algorithms are compared with other related target tracking algorithms in public image dataset.By analyzing the experimental results,the conclusion is drawn: in response to illumination changes,occlusion,deformation,the similarity of target and background and noise interference,the proposed algorithms improves the tracking accuracy and robustness to a great extent.To sum up,the proposed algorithms in this paper can greatly enrich the theoretical research of target tracking field and can meet the partial needs of practical applications in the field of computer vision.