Study Of Visual Multiple Object Tracking Algorithm Based On Intuitionistic Fuzzy Clustering

Visual multiple target tracking(VMOT)is a process of tracking multiple targets and maintaining the identities of targets in video sequences.Because of the significant improvements in object detection techniques in the recent years,many researches in visual multiple target tracking have focused on the tracking-by-detection framework which transforms the multiple target tracking task into a data association problem.Tracking of multiple targets is very important for many real applications.However,in the case of complex background,dense clutter circumstance and high frequency occlusions,how to realize a high-speed and robust tracking of multiple targets is still a big challenge.Therefore,in this paper,a novel data association and trajectory association is proposed for visual multi-object tracking in video scenarios.The main research contents are as follows:For the uncertainty of target occlusion and background interference in single camera multitarget tracking system,a new method of data association based on maximum entropy intuitionistic fuzzy clustering is proposed.First of all,in the proposed algorithm,the affinity of multiple features between the objects and the detections are calculated.Then,the neighborhood rough set theory is applied in feature selection algorithm of object tracking and it can have the desirable effect of reducing redundant features.The fusion of selected features and the local information of the target and observation into eigenvectors constitute the distance between the targets and the measurements.Besides,the association costs between targets and measurements are replaced by the intuitionistic fuzzy membership degrees which are obtained by a modified maximum entropy intuitionistic fuzzy clustering.In addition,Kalman filters is used to estimate the target location.The experimental results show that the proposed algorithm,which has strong robustness and accuracy,can continuously and effectively track multiple targets on multiple video data sets.The proposed algorithm outperforms most existing algorithms in the multitarget tracking accuracy and the number of missed inspections.For the problem of target trajectory interruption and false observation of multi-object tracking with long-term and high frequency occlusion,a fuzzy trajectory association algorithm based on trajectory confidence is proposed.Firstly,in the algorithm,target trajectory confidence and the affinity of the trajectory segment are defined.Secondly,use the threshold to decompose a multi-object tracking problem into high confidence target trajectory and low confidence target trajectory.Then,we give priority to the correlation of the high confidence target trajectory and the observation,and use the obtained correlation results to interconnect the low confidence target trajectory.Finally,in order to reduce the influence of false observations on association,a new trajectory initiation principle and trajectory termination principle are proposed.The experimental results demonstrate that the proposed algorithm can connect the interrupted target trajectory into a complete target trajectory,effectively reduce the number of target tag changes,and improve the accuracy of multi-target tracking.