Moving Object Tracking Based On Multi-independent Features Distribution Fields

Moving object tracking is one of the important research problems in the field of computer vision.And the moving object tracking technique has been widely used in various aspects,such as video surveillance,motion analysis,vehicle navigation,and human–computer interaction.In recent years,with the development of medical technology,financial information,military technology and the demand for them,more attention have been focused on the research of moving object tracking.Many scholars' researches and analyses on the target modeling,feature representation,template updates and other raised the tracking accuracy and robustness to some extent.However,it's still a severe challenge to establish an effective tracking model to suit for more sophisticated scenes.In this paper,multidimensional information is abstracted to design the tracking method according to current scholars' object representation theory,the mainly as follows:Elaborate the relevant theories and references,which mainly start from the domestic and international trends in this field and the important related theories.All above are used to make introduction and analyses for current developments.Proposed the multi-independent features distribution fields(MIFDFs)tracking method to the key problems that object representation and object location.T here are three main steps to implement the method.T he first step is to introduce the distribution fields model,and use the specific layer number that derived from pixel 's feature value to substitute the constant value one,t hereby establishing a multi-layer feature space model.The second step is to extract the comprehensive spatial feature similarity(CSFS)from the multi-layer feature space model.The above two steps not only retain as much of the physical information,but also make full use of the pixels ' information to provide a more accurate weight to indicate pixels ' contribution.T he interaction between the above two make the more accurate object representation.During the third step,on the one hand,the distribution fields is extened to multi-feature space,on the other hand by the means of the CSFS to achieve the moving object tracking result under the condition that the two feature are independent but complementary,both work of this step relized the multi-independent feature distribution fields moving object tracking method.The proposed method improve the original distribution fllieds to achieve the multilayer feature space model,and extract the spatial information,which provide accurate information for the image matching during the tracking process.Besiades,extent the distribution fields to multiple features,which work independent but have complementary perfoamnce,and it is more reliable for accurate object positioning,and it also decrease tracking drift further.In order to verify the contribution that made by the proposed method,we apply it to the target block,rapid deformation,deformation mix,scene chan ges,etc.T he experimental results show that the MIFDFs can outperform other techniques in tracking robustness and accuracy,by the data analyzing,it especially exhibits significant improvement in terms of tracking drift in complex scenes.