Pedestrian re-identification method and system based on unsupervised cross visual angle metric learning

A pedestrian re-identification and cross-perspective technology, applied in character and pattern recognition, instruments, biological neural network models, etc., can solve the problems of difficult pedestrian accurate modeling, time-consuming and labor-intensive modeling process, saving manpower and material resources, and improving accuracy. , the effect of interference enhancement

CN110135294AInactive Publication Date: 2019-08-16XI'AN INST OF OPTICS & FINE MECHANICS - CHINESE ACAD OF SCI
2 Cites 2 Cited by

Patent Information

Authority / Receiving Office
CN · China
Current Assignee / Owner
Publication Date
2019-08-16
Estimated Expiration
Not applicable · inactive patent

Smart Images

  • Figure 1
    Figure 1
  • Figure 2
    Figure 2
  • Figure 3
    Figure 3
Patent Text Reader

Abstract

The invention discloses a pedestrian re-identification method and system based on unsupervised cross visual angle metric learning, and mainly solves the problems that accurate modeling of pedestriansis difficult and a traditional metric learning method needs a large amount of supervised information. The method comprises the following implementation steps: (1) carrying out feature extraction; (2)constructing a common and characteristic projection matrix; (3) carrying out unsupervised cross visual angle measurement learning; (4) constraining consistent distribution of downlink person characteristics of different cameras; (5) optimizing a projection matrix; and (6) projecting the query samples and the test sample set to a potential feature space, and comparing the feature distances to obtain a sorting result of the test sample set. According to the method, pedestrian characteristics are decomposed into common characteristics and characteristic characteristics, so that the pedestrian characteristic expression precision is improved. Meanwhile, the provided unsupervised metric learning method can better meet the actual requirements of a large number of monitoring networks at present, and the method can be applied to the fields of intelligent monitoring, traffic control, criminal investigation assistance and the like.
Need to check novelty before this filing date? Find Prior Art

Description

technical field

[0001] The invention belongs to the technical field of information processing, and in particular relates to a pedestrian re-identification technology in a monitoring scene, which can be used in the fields of public safety intelligent monitoring, traffic control, criminal investigation assistance and the like. Background technique

[0002] With the development of monitoring technology, more and more cameras are used in security systems. Pedestrian re-identification is a technology to find the same pedestrian from surveillance cameras that do not overlap in space. In a large-scale surveillance network, person re-identification technology is very important for target tracking and behavior analysis. Due to the difference in imaging conditions, the apparent characteristics of pedestrians under different cameras vary greatly, such as brightness, posture, occlusion, etc., which brings great challenges to pedestrian re-identification.

[0003] At present, researche...

Examples

Embodiment Construction

[0052] Below in conjunction with accompanying drawing and specific embodiment the step that the present invention realizes is described in further detail:

[0053] refer to figure 1 , the steps that the present invention realizes are as follows:

[0054] Step 1. Obtain image data from multiple cameras that do not overlap in space, and construct a training set.

[0055] Step 2, feature extraction of pedestrian images in the training set;

[0056] (2a), use the marked pedestrian data outside the training set to train the convolutional neural network.

[0057] (2b). On the training set, use the trained convolutional neural network to extract pedestrian feature expressions for each image.

[0058] Step 3. Construct a common and characteristic projection matrix to obtain the final pedestrian feature expression;

[0059] (3a), common projection matrix U 0 : The commonality projection matrix is ​​used to extract common features between all cameras. For the i-th pedestrian image...