| Object tracking is to locate the position of object in the consecutive image sequence which based on the object features(such as contour,size,texture or speed)in the current frame and the historical frame.Object tracking can provide the trajectory,appearance,behavior features of the object which is the basis information for high-level machine vision processing,such as the behavior analysis and object recognition.This paper mainly focuses on how to track multiple moving targets under the changing background and uses correlation filters and FPGA technology to achieve the purpose of real-time online tracking.The correlation filter template is easily to be polluted by the influence of background change,illumination change,target scale change and occlusion which cause tracking failure.In face of this problem,start with target modeling,a correlation filter tracking algorithm based on multi feature fusion is proposed.This method combines gradient feature and color feature,dynamically adjusts the learning rate of target tracking template,takes the similarity of correlation filter response and Gaussian distribution as the judging standard.It can effectively relief the degradation of the target template and solve the problem that one single feature can't resist the influence of fast moving light changing and background.Then based on hue projection,a patch track model is proposed.This method uses spatial structure information to block the target and cooperates with each sub patch to resist the interference of the target being partially occluded.On this basis,according to the location information if each patch estimates the change of object scale,an occlusion and scale changing resist patch correlation filter tracking algorithm is designed.Finally,combining two correlation filtering tracking methods the real-time tracking of multiple moving objects is realized.In order to achieve the purpose of real-time online tracking,realize the key algorithms on FPGA with preliminary simulation and test. |
PDF Full Text Request