Consistency constraint feature learning-based pedestrian re-identification method

Abstract

The invention discloses a consistency constraint feature learning-based pedestrian re-identification method. The method includes the following steps that: S1, pedestrian images are obtained through a camera network, and a training set is marked, parameters are set, and a convolutional neural network is initialized; S2, a picture subset is sampled from a database, the convolutional neural network is adopted to extract feature information, and the similarity matrix of all pedestrians is calculated according to the feature information; S3, the optimal match of the relation matrix of all the pedestrians is solved according to a preset objective function and a preset gradient search method; S4, a gradient backward result is solved according to the difference of the optimal match of the relation matrix of all the pedestrians and a relation matrix of the all pedestrians which is obtained according to actual conditions, and the convolutional neural network is trained according to the gradient backward result; and S5, the step S2 to the step S4 are repeated until the requirements of a user are satisfied. The consistency constraint feature learning-based pedestrian re-recognition method of the invention is applicable to a condition in which matching is performed under a large-scale camera network and eliminating conflicting matching errors.

Description

technical field [0001] The invention relates to the technical field of digital image processing, in particular to a pedestrian re-identification method based on consistency constraint feature learning. Background technique [0002] Person Re-Identification (Person Re-Identification) is to match the collected pedestrians under the perspective of different cameras, and judge whether different pictures belong to the same person. Pedestrian re-identification has a wide range of applications and broad prospects in monitoring security and other fields. However, because the collected pedestrian pictures have great changes in size, illumination, viewing angle, posture, etc., although many researchers have participated in recent years It has not been well resolved in related researches. [0003] The current pedestrian re-identification methods are mainly based on pairwise re-identification, that is, each time only two collected pictures are considered whether they belong to the same...

AI Technical Summary

Problems solved by technology

[0009] The above methods are based on a pair of cameras for matching, which is quite inconsistent with the camera network environment composed of hundreds of cameras in real life.

At the same time, all methods use the pairwise comparison method, only considering two pictures each time, so that the structural features under the camera are not well used to help match, and there will be conflicting matching situations, such as figure 1 Among them is an example, where pedestrian P1 and P2, P1 and P3 are considered to be the same person, while P2 and P3 are considered to be different people, resulting in low pedestrian recognition accuracy

Images