Research On Methods Of Multiple Object Tracking Based On TSK Fuzzy System

Multiple Object Tracking(MOT)technology plays a key role in the fields of computer vision,intelligent surveillance,pattern recognition andother fields,which has attracted more and more scholars' attention.Although many methods for video multi-object tracking have been proposed,in complex scene,accurate tracking of targets over long periods of time remains a challenging topic.There are certain indeterminacy factors in appearance,including illumination changes,clutter interference,and target scale changes;some elements during the tracking process,such as unknown object number,objectmissing,and false observation.Due to these reasons,object tracking become difficult.Aiming at the uncertainty of video object tracking in complex environment,a video object tracking algorithm based on TSK fuzzy system is proposed.In this algorithm,firstly,the similarity between the object and the observation is extracted,then input into the constructed TSK fuzzy semantic model,the similarity is weighted and fused by the model,and the membership matrix between the target and the observation is obtained,which can achieve the connection between the object and the observation;After getting the stable track of the object by algorithm based on TSK fuzzy semantic model,we will use the multi-frame information to train the TSK fuzzy multi-classifier system.Each target with the stable track will have the trained TSK fuzzy classifier model.When the observed features are input into the classifier model,the observed label matrix can be obtained,which can be used to correlate the observation with the object with stable track.Through tracking effect and data analysis,the proposed algorithm can obtainaccurate tracking results in complex environment.In order to make better use of the useful information of uncertain information in video multi-object tracking,a data association algorithm based on multi-feature TSK fuzzy classifier is proposed.In the process of training,the algorithm adds cooperative learning mechanism,and makes effective use of the independent information of each feature and the correlation information between each feature.The TSK fuzzy classifier based on the learning mechanism can better realize the data association between the target and the observation.Theexperimental results show that the algorithm can achieve good tracking performance on multiple test videos.