Research On Pedestrian Re-identification Method Based On The Constraints Of Perspective And Label Consistency

Pedestrian re-identification is to match people in non-overlapping camera images.This is a very challenging research topic because of the great changes in posture,lighting,and the switching of perspectives.How can such a complex background be used in surveillance camera networks? Matching to the same pedestrian not only requires high robustness of the feature,but also requires applicability and generalization of the algorithm.Person re-identification is to match persons across non-overlapping camera views.It is a very challenging research topic due to large changes in pose,illumination and viewpoint.To improve the robustness to such variations,we develop a joint asymmetric projection and dictionary learning algorithm by adopting listwise similarity and identity consistency constraints.Benefiting from the introduction of listwise similarities,the similarity list between each pedestrian image is considered during learning the dictionary,thus,a large amount of discriminative information contained in samples is effectively exploited.That endows the dictionary with discrimination.In addition,we impose a identity consistency constraint on the coding coefficients to further improve the discriminative ability of the dictionary.To alleviate the detrimental effect on pedestrian matching that arise from the appearance changes across non-overlapping camera views,two asymmetric projection dictionaries are employed to map the pedestrian features into a unified subspace such that the correlation between data from the same people in different views is maximized.Finally,by integrating the coding coefficient and classification results,we develop a fusion strategy with a modified cosine similarity measure to match the pedestrians.Experimental results on different challenging datasets demonstrate that our method is effective and outperform some state-of-the-art approaches.The self-encoding dictionary has good discrimination compared to other dictionary learning methods.Specifically,it combines the comprehensive dictionary and the analysis dictionary to learn at the same time,and adopts the mechanism of analysis and coding.By further improving the construction of the self-encoding dictionary,making it a positive definite matrix,it is guaranteed that the local optimal solution is the global optimal solution.From the view point of view,the Pair-learning dictionary learning is used in pedestrian re-recognition to train a comprehensive dictionary and perspectivebased analysis dictionary,which not only reduces the computational pressure on the algorithm complexity,but also compares the recognition rate with other methods.More improved experimental results.We use the convolution feature to improve the robustness of the features from their perspectives,and use the method of Center Loss to enhance the distance in the face recognition,and combine the loss function and convolution features of the pedestrian identity layer to train.Experiments have shown that Center Loss and identity loss functions based on two perspectives achieve better results than most methods.In the end,we summarize the future trends and existing problems in research and development our ideas into practice.I will not only study the latest research results,but also contribute a meagre strength to research field.