Group Of Robot Visual Based On Diffusion Distance Measure To Target Re-identification

In recent years,in special scenarios such as the disaster,through the robot vision to screening and disaster rescue is playing an increasingly important role.Which involve the most critical vision technology---target re-identification.Target re-identification,that is,to judge images collected from different robot of pedestrians whether are the same person.However,due to the difference of illumination and view angle,lead to the different images of physical characteristics have great differences even though the same pedestrian.Based on the above problems,this paper on the basic of robustness characteristics' expression,mainly carry out research from measurement algorithm,specific work is as follows:In this paper,we established a similarity measurement function based on the optimal diffusion distance for the target re-identification.To improve the computation efficiency of original algorithm,in the high dimensional diffusion distance of the expression,to join the Gaussian pyramid two-thirds ratio of image sampling;at the same time,to improve the matching precision of original algorithm,considering the differences and similarity of samples to weighted processing.Against traditional KISSME algorithm,the problem existing in the small sample characteristic value of instability,and once the new sample entered to training,which lead lowly computation efficiency.So,adaptive delta KISSME algorithm is proposed.Optimization is divided into two steps,in order to solve the problem of the eigenvalue,to join the virtual sample,smooth and regularization technique,thus increasing sample size and inhibition of instability of the characteristic value;In order to improve the calculation efficiency,in view of the new sample to join adaptive incremental learning,avoiding the need to burden of training time once new samples entered.Finally through experiments show that the proposed measure method can good achieve target re-identification,has obvious advantages compared with other algorithm.Due to realistic scene demand,the paper of proposed method of key link,to promote the field further realize the professional,efficient,intelligent,has far-reaching significance.