A Research On Object Tracking Methods Based On State Estimation

In recent years,with the rapid development of the monitoring network,object tracking already has a wide application prospect in military,medical,traffic system and public security,etc.Object detection and tracking is an important way to keep obtaining useful information in the video.According to extracting effective extracted feature of object,it can gets the real-time state of the target.The traditional object tracking algorithms are difficult to capture the position information and reduce tracking performance in case of the target occlusion and the minimal target.In this paper,we study the target on the condition of occlusion and the minimal target.The main achievements include: firstly,we use the sub-regions tracking model,weighted clustering algorithm,Kalman algorithm and thento accurately track the occluded object accurately;Secondly,we use the strategy of sub-region-resampling and the method of trajectory similarity to promoting promote efficiency of partially-occluded object tracking.Finally,we comprehensively use of the underling characteristics of the target and improved Kalman filter to implement minimal targets tracking in unmanned aerial vehicle(UAV)scene video.The experimental results shows that the proposed methods has have obtained a goodbetter performance than that of tradition.The specific work is summarized as follows: 1.The tracking method of the occluded objects has been studied thoroughly.Aiming at the problem of long-term tracking algorithm,TLD will be tracking drifting or false when tracked targets encounter obstacles occlusion,we proposed an occluded object tracking method based on sub-region clustering and Kalman filter.This method judges the occlusion of the target based onaccording to the mutual information between objects.When the target is occluded partially,this method introduces sub-region-TLD tracking model and unsupervised learning to track the target accurately and avoid tracking drifting.When the target is occluded seriously,this method introduces linear Kalman filter method to effectively estimate the target position.2.The efficiency of sub-region clustering method in this paper has been studied thoroughly.Aiming at the low efficiency of the partially-occluded object tracking method based on subregion clustering,we proposed a partially-occluded object tracking method based on sub-region resampling.This method uses resampling model in particle filter and trajectory similarity theory.It keeps high precision and accuracy and improved tracking efficiency.3.The method of minimum target tracking in UAV scene video has been studies.General tracking methods is difficult to tracking minimum target accurately.We proposed a minimum target tracking method based on improved Kalman filter.This method utilizes detection results as observations,and uses similarity measures tracking error to determine covariance matrix of observation noise.We choose different tracking model depending on cumulate error of detection results and tracking results,to establish a complete minimum target tracking framework.