The Single Objective Multi.scale Tracking Algorithm Based On Compression Perception Research

Video tracking in video surveillance,human computer interaction, traffic monitoring and other fields has important applications, although in these areas has achieved some success, but due to illumination change, target deformation and acceleration factors such as interference, to establish a stable, high efficiency of tracking system is still very challenging.In this article, we proposed two kinds of robust real.time tracking algorithm, used to solve encountered in the process of tracking of interference.Based on the prediction of multi.scale compression tracking.In the tracking algorithm based on detection, target said is apparent model was constructed by target feature extracting.And in order to reduce the amount of calculation, improve the real.time performance, the traditional way, have a plenty of subspace model was constructed to represent the target, have a plenty of use increase quantum space model to represent the target to adjust to changes in the target appearance, some by support vector machine(SVM) to improve the choice of discriminant features.In order to reduce the amount of calculation algorithm, we proposed the multi.scale compression tracking algorithm based on prediction.The algorithm to extract positive samples in the prediction of target location near, near away from the location of the prediction to extract negative samples, and convolution with multi.scale box filter to target the multi.scale feature vector.The position of the forecast with the actual location close to target, so, in the search target characteristics, adopted by the search radius don't need too big, effectively reduce the amount of calculation.However, as a result of building multi.scale high.dimensional feature vector dimension is bigger, still need a lot of calculation in the process of tracking, using compressed sensing technology to reduce the dimensions of the feature vector, to reduce the amount of calculation in the process of tracking, and basic retain the original characteristics after the dimension reduction feature vector most of the information.After the use of compression perception for dimension reduction of compressed domain, naive bayesian classifier is used to analyse the characteristics of the classification, the classification of the greatest feature of the mark position, target tracking location.Multi.scale compression tracking algorithm based on color correction.Alt.ho ugh the compression tracking algorithm based on the detection and the improve d algorithm realized the real.time,extracting the single tracking target feature,e xtracting only the gray haar.like the characteristics of the tracking object, cause d the construction of a model of apparent when tracking robustness is not very good.So,is proposed in this paper, based on the multi.scale compression of the color correction tracking algorithm.On the basis of the original algorithm, com bined with color information of the target, and combined with particle filter fra mework, implements the algorithm of multi.scale tracking.Also made improvem ent on classifier update methods, using double s.shaped curve, the classifier par ameters to realize the adaptive update, better suppress noise interference and th e classification of error caused by the tracking failure.Thesis experiment is in the Windows7 platform, using MATLAB2014 a to programming.In the experiments, used to detect the test sequence from the benchmark test standard library provides.Comparing algorithm code supplied with the author,parameter Settings are the default Settings.Through experiment, it shows that multi.scale compression tracking algorithm based on prediction to illumination change, deformation, shelter, and background clutter situation has better robustness,compared with CT, FCT has grown, MSCT tracking algorithm based on compression perception, faster and better robustness;Multi.scale compression tracking algorithm based on color correction, in a complex scene, for deformation occurred in the process of target tracking, shade and dramatic lighting change situation stability, robustness and accuracy are better, reached the expected requirement.