Multi-shot pedestrian re-identification method based on sparse representation

Abstract

The invention discloses a multi-shot pedestrian re-identification method based on sparse representation. K-means clustering algorithm is adopted to cluster all GOG eigenvectors in each eigenvector setcorresponding to the first image library into a plurality of clusters; all the GOG eigenvectors in each feature vector set corresponding to the second image library are clustered into multiple clusters, and the mean value feature vectors of all the GOG feature vectors in each cluster are used as corresponding feature representations, so that redundant information is eliminated, and the pedestrianre-identification speed is increased; meanwhile, a strategy of learning an embedded space by combining a kernel local Fisher discrimination algorithm is combined, so that the influence of illumination change and view angle change on pedestrian re-identification is further eliminated, and the accuracy of pedestrian re-identification is improved; the correlation information between the pedestrian images of the same pedestrian in the first image library and the correlation information between the pedestrian images of the same pedestrian in the second image library can be effectively utilized atthe same time, so that the constructed joint group sparse representation model has higher robustness on object shielding and human body posture change.

Description

technical field [0001] The present invention relates to an image recognition technology, in particular to a sparse representation-based Multi-shot pedestrian re-identification method. Background technique [0002] Given an image of a person of interest in one camera view (called the test view), the goal of person re-id is to automatically re-identify the same person from another, disjoint camera view (called the target view). Nowadays, many high-level applications in computer vision rely on accurate pedestrian recognition results, such as object tracking, intelligent video surveillance, etc. Pedestrian re-identification has always been one of the most challenging tasks in the field of computer vision because pedestrian images are easily affected by factors such as illumination changes, human body posture changes, and occlusions. How to accurately re-identify pedestrians has been deeply researched by relevant institutions at home and abroad. [0003] At present, the researc...

AI Technical Summary

Problems solved by technology

Such as: S.Karanam, Y.Li, and R.J.Radke.Sparse re-id: Block sparsity for person re-identification[C] / / Computer Vision and Pattern Recognition(CVPR), 2015:33-40.(Karanan Si Rikrishna, Li Yang, Radek Richard, Pedestrian Re-Identification Based on Block Sparse Representation, Computer Vision and Pattern Recognition, 2015:33-40), which proposed a block sparse representation method, whose By defining all images in the same pedestrian sequence in the target set as a group, the block sparse representation does not expect the sparse coefficients to have as few non-zero elements as possible, but considers that the non-zero coefficients should be concentrated in a specific group, however , although the block sparse representation method has considered the correlation information between the images in the same pedestrian image sequence in the target set, it still sparsely encodes each image to be tested one by one, ignoring the individual Correlation information between images (since the images in the same pedestrian sequence in the image library to be tested are relatively similar, there is some complementary information between them)

Images