Research On Person Re-identification Based On Dictionary Learning And Fisher Discrimination Sparse Representation

Due to the popularity of video surveillance systems,cameras are arranged in a variety of public places,providing a large number of surveillance videos for pedestrian tracking.The way to manually monitor and analyze video consumes a lot of manpower and material resources.Second,it is slow and inefficient.Therefore,intelligent analysis video method based on pedestrian recognition is becoming an urgent requirement.Person re-identification is applied widely in the areas of video surveillance.It aims to identify all images of one pedestrian taken in different scenes from different cameras.As the scene changes led to the light,shooting perspective changes make pedestrians appearance of pedestrian images vary widely in different scenes,which brings some difficulties to pedestrian recognition.Dictionary learning can better express image features with different pose,illumination fluctuations,object occlusion and scale change,it is an important person re-identification method and can achieve higher recognition rate.However,the present dictionary learning methods failed to fully consider the connection of the pedestrian features from different camera views.For this reason,this paper proposes a new method based on dictionary learning and Fisher discriminant sparse representationBy considering the features of the same pedestrian in different scenes should have similar sparse representation,this paper applying the Fisher discriminant criterion to the traditional dictionary learning framework,and putting forward the concept of person re-identification scatter function through adding a regularization term which constrains the sparse representation.The regularization term is aiming at maximizing the between-class scatter of the sparse representation of different pedestrian,and minimizing the within-class scatter of the sparse representation of same pedestrian.With the learned dictionary sparse representations were achieved which are more discriminative.And we design experiments on international public datasets,including VIPeR dataset,PRID450 s dataset and CAVIAR4 REID dataset.The experimental results show that the recognition rate of the proposed method is higher than the person re-identification methods based on dictionary learning.And verify the performance of our dictionary learning framework in improve the recognition rate of datasets.In addition,this paper discusses the application of the deep learning method in personre-identification,and adds the deep learning to the dictionary learning framework,proposes a new person re-identification method based on deep learning and dictionary learning.That is,through the defined deep convolution network structure to extract pedestrian image characteristics as the features used by dictionary learning method.The experiment results indicate that the dictionary learning framework proposed in this paper can still maintain the advantages in the recognition rate of image dataset,which is based on the depth feature.