Pedestrian re-identification method based on unsupervised local measurement learning and reordering

A pedestrian re-identification and metric learning technology, applied in the field of pedestrian re-identification based on unsupervised local metric learning and reordering, can solve the problems of limited training data and large differences between different pedestrians, and achieve low time complexity, feasibility and The effect of improving usability, enhancing usability

Active Publication Date: 2017-12-22
UNIVERSITY OF CHINESE ACADEMY OF SCIENCES
View PDF3 Cites 23 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

Due to the limited training data, there are huge differences between different pedestrians in the actual scene, and the global

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Pedestrian re-identification method based on unsupervised local measurement learning and reordering
  • Pedestrian re-identification method based on unsupervised local measurement learning and reordering
  • Pedestrian re-identification method based on unsupervised local measurement learning and reordering

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0090] In this implementation, it is only necessary to obtain the local metric matrix M corresponding to the sample in the image library i , to calculate the similarity.

[0091] In a further preferred embodiment, query samples x are used respectively 0 The local metric matrix M of and the sample g in the image library i The corresponding local metric matrix M i , calculate the similarity between the query sample and the sample in the image library, and add the two similarities to get the final similarity.

[0092] The similarity is expressed in the form of distance as follows:

[0093] d(x 0 , g i ) 2 =(g i -x 0 ) T (M 0 +M i )(g i -x 0 ) Formula (4-3).

[0094]Through formula (4-1), formula (4-2), and formula (4-3), the distance between the query sample and the sample in the picture library can be calculated, and the similarity ranking can be obtained. The shorter the distance, the higher the ranking and the greater the similarity.

[0095] Using the local me...

Embodiment 1

[0115] 1. Database and sample classification

[0116] The pedestrian re-identification detection is carried out by adopting the method of the invention. For the accuracy and comparability of the experiment, the widely used public data VIPeR, CUHK01 and PRID2011 databases in the field of person re-identification are used.

[0117] VIPeR dataset: It consists of 1264 pictures of 632 people under two cameras, each person has only one picture under each camera, and the pictures are normalized to 128*48 pixel values. In addition to the different viewing angles of the two cameras in this dataset, the lighting conditions vary greatly, which brings great difficulty to re-identification. In the experiment, we use 316 samples under camera a as the training sample set, the remaining 316 samples under camera a as query samples, and 316 samples under camera b corresponding to the query samples as the image library.

[0118] CUHK01 dataset: Contains a total of 971 people, each with two ima...

Embodiment 2

[0131] 1. Database and sample classification

[0132] The VIPeR, CUHK01 and PRID2011 databases were used for experiments. The VIPeR and CUHK01 data sets are divided on the basis of Example 1, and the number of query samples is reduced to half of the original, respectively 158 and 243 query samples, and the division of the training sample set remains unchanged.

[0133] In the PRID2011 data set: 100 samples from 200 pairs of cameras a are randomly selected as query samples, the remaining 100 samples from camera a are used as training sample sets, and 649 samples (100+549) from all remaining cameras b are used. as a photo gallery. At this time, there are many negative samples in the image library, and the interference ability is strong.

[0134] 2. Performance evaluation criteria

[0135] Make a CMC curve and consider the matching accuracy at rank-1 as an index to measure the effectiveness of the method.

[0136] For each query sample and sample in the picture library, calcu...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

PUM

No PUM Login to view more

Abstract

The invention discloses a pedestrian re-identification method based on unsupervised local measurement learning and reordering. The method comprises the steps that a pedestrian picture is acquired; a query sample is determined, and a training sample set and a picture library are formed; feature extraction is carried out on the acquired pedestrian picture, and features are described as eigenvectors; a corresponding measurement matrix is acquired by learning local measurement for each sample in the query sample and/or picture library; similarity calculation is carried out through the measurement matrix acquired through learning, and initial ranking is carried out according to the similarity magnitude; and initial ranking is optimized through re-ranking to acquire a final ranking result. The method is based on unsupervised local measurement learning, is free of manual sample annotation, and has the advantages of practicality and scalability. Through re-ranking, the matching accuracy is further improved.

Description

technical field [0001] The invention relates to the fields of computer vision and image processing, in particular to a pedestrian re-identification method based on unsupervised local metric learning and reordering that can be used in the fields of intelligent video surveillance and the like. Background technique [0002] Cross-camera pedestrian re-identification is currently a hot research issue in the field of intelligent video surveillance. Its main purpose is to obtain the action trajectory of a specific target in a specific time in a specific camera network coverage area. This is of great significance in the context of the current big data era, video surveillance automation, and safe city construction. With the maturity of monitoring equipment related technologies and the reduction of costs, tens of thousands of monitoring equipment are collecting data all the time, forming a massive database. How to effectively manage and utilize monitoring big data has become a widely...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to view more

Application Information

Patent Timeline
no application Login to view more
IPC IPC(8): G06K9/00G06K9/46G06K9/62
CPCG06V40/103G06V10/467G06V10/50G06F18/2155G06F18/2411G06F18/22G06F18/24147
Inventor 韩振军赵恒叶齐祥焦建彬
Owner UNIVERSITY OF CHINESE ACADEMY OF SCIENCES
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap