Research On Methods Of Visual Object Tracking Based On Fuzzy Information Processing

With the development of computer technology and sensor technology,the technology of video capturing is continuous improving,and the amount of videos which are collected everyday are also increasing.The requirement of processing massive video data promotes the rapid development of intelligent video analysis technology.As one of the core technologies of intelligent video analysis,visual object tracking can estimate the states of the targets,such as position,label,size,and speed,and reconstruct the trajectories,which provides important basis for high-level tasks such as behavior analysis and scene understanding.Visual object tracking has become one of the research hotspots among domestic and foreign scholars and research institutions.Currently,visual object tracking has been widely used in video surveillance,intelligent transportation,intelligent robots,human-computer interaction,video compression,video retrieval,unmanned driving and many other areas,it has large research significance and academic value.Although many visual object tracking methods have been proposed,long term robust object tracking in a complex scene is still a challenging problem.Due to the lack of prior knowledge and the presence of illumination changes,background clutter,scale variation and many other factors during the tracking process,it is inevitable that the appearance feature of the object contain uncertain information which makes tracking the target become difficult.For multiple object tracking,many other uncertain factors,such as the unknown of the number of the targets,miss detection and false positive,will add more difficulties in the data association between the targets and the observations.Focusing on how to address the uncertainty in the appearance model and the data assoicaition,the visual target tracking problems are deeply studied base on the theory of fuzzy set and intuitionistic fuzzy set.A series of effective visual object tracking methods are proposed as follows:Firstly,for the problem of tracking single object in the complex scene,a multiple attribution fuzzy synthetic based visual object tracking algorithm is proposed.In the proposed algorithm,the appearance feature and the optical flow feature are integrated into a multiple feature voting framework for tracking.Meanwhile,in order to handle the uncertainty both in the voting confidence of appearance feature and in the voting confidence of optical flow feature,the fuzzy synthetic function is introduced to integrate the voting confidences which can produce reliable fuzzy synthetic confidences.Then,the position of the center of the object is estimated based on the fuzzy synthetic confidences.Experimental results show that the algorithm can achieve better tracking results even under the conditions of illumination change,background clutter and scale variation.Secondly,in order to handle the problem of data association in visual multiple object tracking under the conditions of many uncertain factors,such as the inaccurate locations,the varing number of targets and the false positives and so on,a fuzzy spatial and temporal information clustering based multiple object tracking algorithm is proposed.In the proposed algorithm,multiple features are extracted from the spatial and temporal information.The fuzzy c-means clustering are carried out twice to calculate the fuzzy membership degrees of using the targets as the cluster center and the fuzzy membership degrees of using the observations as the cluster center,respectively.Then,the two kinds of fuzzy membership degrees are fused by the defined fuzzy synthetic function.The data association is realized based on the fuzzy synthetic membership degrees.Experimental results show that the algorithm can effectively deal with the uncertain factors in data association,meanwhile,suppress the false tracks and achieve good tracking results.Thirdly,in order to better extract the useful information in the uncertainty in data association of multiple object tracking,an intuitionistic fuzzy set based multiple object online tracking algorithm is proposed.In the proposed algorithm,the intuitionistic index is defined to model the uncertainty in the fuzzy membership degrees obtained from fuzzy c-means clustering.Then,the useful information is extracted from the intuitionistic index through intuitionistic fuzzy point operator,and the intuitionistic fuzzy membership degrees are calculated.The data association is realized based on the intuitionistic fuzzy membership degrees.Experimental results show that the algorithm can better handle the uncertainty in data association of multiple object tracking,and achieve good tracking performance in several public test videos.Fourthly,in order to handle the uncertainty in the description of feature in visual multiple object tracking,an intuitionistic fuzzy random forest based multiple object tracking algorithm is proposed.In the algorithm,a novel intuitionistic fuzzy decision tree model is designed and used as the base classifier.Then,combined with random sampling and random feature selection,an intuitionistic fuzzy random forest model is given which is used as the classifier to track the missing targets.Experimental results show that the proposed algorithm achieve improvements in many evaluation metrics such as multiple object tracking accuracy,compared with other tracking algorithms.In addition,considering that the intuitionistic fuzzy random forest needs many storage and computing resources,a boosting intuitionistic fuzzy tree model is proposed which is built by boosting the shallow intuitionistic fuzzy decision tree model with the Adaboost algorithm.