Pedstrian Target Recognition Across Cameras

In large camera monitoring tracking network,Due to the too munch uncertainty of tracked target,such as monitoring perspective,environment light and target attitude change.How to identify the target is a difficulty.To Extract a robust statistical appearance features of the target is the core of the recognition problem .This paper put forward a target recognition method,which based on attention feature matching.First,take from the local color texture feature(ColorSIFT) from target image in the camera.Then,according to the color of the local texture feature we define a search criteria to generate attention significant figure of the target image.Finally combined with the feature of significant figure and ColorSIFT,we use a two-way identification to finish the target recognition.Paper's main work and innovation are as follows:(1) Through the analysis of a lot of research work both at home and abroad,and summarizes multiple viewpoints of tracking problem.And find out the key point is that how to establish a model of target appearance,which can make it remain stable or small changes in appearance in the cameras network,and how to measures this model's similarity.The goal of this paper introduces several common appearance model and similarity measurement,we summarizes the current research status at home and abroad,and puts forward a method which first correct the differences of camera imaging,and then by extracting features of visual attention and target recognition.(2) Due to the imaging parameters,camera angles,scene lighting,and other aspects,the influence of the target imaging differences in multiple cameras.Color features as an important clue in the target recognition problem is affected by the enormous influence.This paper introduced a brightness transfer function (BTF) concept,through the study of the transfer model of color space between multiple cameras,it decrease differences imaging of the cameras in multiple cameras.It can lay a good foundation to extract target feature and generate attention.(3) Through the analysis of visual attention,we find the characteristics that human eye vision always can quickly find the interested target.This paper put forward a target recognition method based on attention feature matching.First extract the local color texture feature(ColorSIFT) from the camera target image,then according to the color of the local texture feature to search under a to certain criteria,and generated attention significant figure,finally combined with the feature of significant figure and ColorSIFT.The experimental results show that,as a result of the attention information to do the identification,This method is not sensitive to the background noise,It can ignore the complexity of target background noise extraction in the process of introducing,at the same time the attention into a stage of target attitude change tolerance is higher,It can effectively overcome the effects of posture change on the recognition,In the real test it's results show that the algorithm by background noise and the influence of object pose,illumination change is small,we have a higher recognition rate.