Research On Moving Target Tracking Technology Based On Complex Background

Target tracking technology,as an basic support technology in machine vision,can be widely applied in the intelligent transportation,medical image,human-computer interaction,mobile camera and other areas.Relevant scholars have carried out research on tracking problems and achieved some results,However,there are less research on background clutter,light change,partial occlusion and other tracking problems under complex background.Therefore,this paper carries out the following research work based on the above problems:Firstly,for the problem of moving target tracking under complex background,the classical algorithm is studied,and the advantages and disadvantages of the algorithm are compared and analyzed based on experiments in this paper,which lays a theoretical and technical foundation for solving the problem of video target tracking in complex background.Secondly,aiming at the problem of poor tracking accuracy caused by light variation and background clutter interference,a particle filter target tracking algorithm based on multi-feature hierarchical fusion is proposed.In this algorithm,motion detection based on frame difference method is added to the particle filter prediction to generate candidate target samples.At the same time,image blocks detected with large similarity are added to the particle population and transmitted to the next generation.When calculating the fusion weights of particles,in order to give full play to each feature's the advantages in dealing with different scenarios,the hierarchical feature is determined firstly according to the descriptive ability of each feature in a specific scenario,and then the particle swarm is layered based on this feature to get the corresponding weight parameters of each particle layer,and then uses the weight parameters and the weight of another feature to get the fusion weights of the two features.After locating the target at a new time,the target template is updated according to a certain proportion of the target information when the update requirement is satisfied.Thirdly,for the problem of poor tracking robustness caused by background clutter interference,target appearance deformation and fast motion,a hybrid target tracking technology based on improved fireworks algorithm is proposed.For eliminating redundant information and preserving color information at the edge contour,when extracting picture feature information a gradient sparse matrix isproposed as a feature information extraction tool to extract color feature information of the picture;At the same time,in order to improve the prediction accuracy of the motion model,an improved fireworks algorithm is proposed to generate candidate target samples.This algorithm is different from the traditional fireworks algorithm,which uses fixed invariant constants as explosion radius parameters,but adds target motion rate to adjust fireworks radius parameter on the basis of it.And for controlling the diversity of fireworks,the algorithm also dynamically adjusts the probability of variation of fireworks according to the variation of variance of confidence degree of fireworks,and keeps the diversity of fireworks within the optimal range.In addition,in order to update the appearance model of the target in time,the number of updated samples is adjusted by the change rate of the confidence of the target,and then using online extreme learning machine to learn the corresponding amount of target information incrementally.Finally,we developed a compatibility and stability client software module on the software platform VS2010 embedded with OpenCV2.4.9 and software interface development tool QT.Experiments show that the tracking system has good tracking performance and can meet realistic needs.