Research On Animial Target Tracking Based On Improved Sparse Representation

Nowadays,the research of target tracking is one of the most advanced directions in the field of artificial intelligence and automation control.Animal tracking means automatic detection and tracking of these animals in dynamic videos,which can provide technological support for the research of animal behavior and movement.At the same time,it has great application values in the protection of wild animals and open animal farms.However,the target tracking technology always can be interfered by a variety of animal complex sporting factors.Drift,loss and other phenomena often happened in the tracking process,all of which led to the failure of tracking.In practical application,as a special kind of moving object,the diversity of animals resulted in a variety of animal movement forms.As well known,aninals' living environment are usually very complicated,it is harder for aninal tracking accurately.In fact,there are lack of particular animal target tracking technologies,and it is significant for the study on animal target tracking.In this paper,the sparse representation theory was introduced for improving the exist target tracking technolgy,and made the animal target tracking more efficient and reliable.The contrubutions of this article are as follows:(1)Aiming at solving the problem of target tracking.The target of animal couldn't be tracked accurately.Sparse representation was introduced into the particle filter framework for target tracking and a sparse cooperative model was proposed in this paper.Firstly,the optimal classification features were extracted for classifier by training the positive template set and negative template set in the discriminant classification model.Secondly,target was weighted by the histogram for the processing speed through unsupervised feature learning and greedy algorithm in the generative model.Then,the discriminant classification model and generative model were cooperated in a collaborative model,and the target was determined by the reconstruction error.Finally,every module was updated independently to mitigate the effects of changes in the appearance of the target.The comparison of the experimental results on the public dataset and the animal dataset showed that the average center location error(pixel)of the model proposed was 7.3 and 17.1 respectively.It was smaller than others,meanwhile the model was robust to noise and real-time.(2)In order to improve the accuracy of target tracking and make full use of sample information,an improved sparse representation ranking method was proposed in this paper.Firstly,the candidate targets were represented by sparse block.Secondly,sparse coefficients and residuals were calculated.Then,the ranks information could be extracted.Subsequently,two rankings were fusion together according to the similarity function.Finally,the template was updated by the tracking results.Experimental results showed that the average error(pixel)was 12.2,which emonstrated the effectiveness of our method,and was better than others.(3)Resulted from the actual application requirements,a prototype system of animal tracking platform was designed.The tracking module and interface had been implemented.