Feature sparse representation multi-dictionary pair learning pedestrian re-identification method

Abstract

According to the method, the difference between different cameras and the difference between different features are considered at the same time, the difference exists between the features of different types of the same image in the actual environment, and even the difference exists between the features of the same type in two different cameras. According to the method, sparse representation is carried out on different features in each camera, so that the same features in the two cameras obtain a coupling dictionary pair, and the coupling dictionaries are utilized to reflect the difference between different cameras and the difference between different features. Through analysis of experimental results, the effectiveness of the algorithm is greatly superior to that of other comparison algorithms, whether a specific pedestrian exists in an image or video sequence or not can be accurately judged, a pedestrian monitoring image is given, the visual limitation of a fixed camera is made up by retrieving the pedestrian image in a cross-device mode, and the method can be combined with a pedestrian detection or pedestrian tracking technology, and can be widely applied to the fields of intelligent video monitoring, intelligent security and the like.

Description

technical field [0001] The invention relates to a surveillance video pedestrian re-identification method, in particular to a feature sparse representation multi-dictionary pair learning pedestrian re-identification method, which belongs to the technical field of surveillance video pedestrian identification. Background technique [0002] Surveillance video pedestrian re-identification is a technology that uses computer vision technology to determine whether a specific pedestrian exists in an image or video sequence. It is a sub-problem of image retrieval. Given a surveillance pedestrian image, retrieve the pedestrian image across devices, aiming at It can make up for the visual limitations of fixed cameras, and can be combined with pedestrian detection / pedestrian tracking technology, and can be widely used in intelligent video surveillance, intelligent security and other fields. Due to the differences between different camera devices, pedestrians have both rigid and flexible ...

AI Technical Summary

Problems solved by technology

[0012] Second, sparse representations have been widely used in image restoration, compressed sensing, and face recognition, but research on multi-view pedestrian re-identification methods is still in its infancy.

However, due to the complexity of pedestrian re-identification in the actual environment, most of the existing technologies are only realized under theoretical conditions. At the same time, some key technologies still lack effective theoretical basis and technical support, and there is still a long way to go before practical application considerable gap;

[0013] Third, there is an obvious difference between pedestrian re-identification and applications such as face recognition and palmprint recognition. Pedestrian re-identification must include at least two cameras with no overlapping fields of view. Images of the same person may exist in multiple cameras at the same time. In one camera, and due to the actual environment of the camera, even the images of the same pedestrian in different cameras will be affected by factors such as illumination, angle and posture, and there will be great changes. In solving the problem of pedestrian re-identification , it is necessary to take these influencing factors into consideration as much as possible. Some pedestrian re-identification methods in the prior art project the data or features in the two cameras into the common distance metric obtained through learning, which not only ignores the distance between the cameras The difference between them, and when the data is insufficient, it is easy to cause overfitting;

[0014] Fourth, although there are some methods in the existing technology that take into account the differences between different cameras, they ignore the differences between different features. Different types of features of the image describe different characteristics of the image, and each feature describes the characteristics of the image. A certain feature, and the disadvantages of a feature can be weakened by the advantages of other features. Although different features are combined into a column vector, because this method does not correctly look at the features contained in different features, the method of combining features Unable to effectively use the discriminative information between different features;

However, when the method based on distance learning is applied to the pedestrian re-identification scene, there are still many problems, such as the small sample problem

Images