Identity recognition method and device

Abstract

The embodiment of the invention provides an identity recognition method and device, and the method comprises the steps: obtaining a to-be-recognized image comprising a test object; extracting multi-modal features of the test object from the to-be-identified image according to a pre-established feature extraction model, wherein the multi-modal features at least comprise face features and human bodyfeatures; comparing the multi-modal features with modal features included in each multi-modal set in at least one multi-modal feature set in a pre-established feature registration library, and determining a target multi-modal set corresponding to the highest comparison score from the at least one multi-modal feature set, wherein each multi-modal feature set in the at least one multi-modal featureset at least fuses face features and human body features; and determining identity information corresponding to a target face feature included in the target multi-modal set, and determining the identity information corresponding to the target face feature as the identity information of the test object.

Description

technical field [0001] The invention relates to the technical field of identification, in particular to an identification method and device. Background technique [0002] Face recognition is a technology for identity recognition based on facial images. Specifically, firstly, feature vectors are extracted from face images, and then the similarity between vectors is calculated through a certain similarity measurement function. The current mainstream solution is based on volume The product neural network CNN is used for feature extraction, and the cosine function is used for similarity calculation. In the general application mode, it is necessary to collect face images of multiple target persons first, then use the features extracted from these face images as a registration library, and then perform feature extraction on the faces in the video, and compare them with the features in the registration library. Compare and identify the target person corresponding to the feature if...

AI Technical Summary

Problems solved by technology

[0003] However, there are problems such as large face angles and occlusions in complex scenes, which will lead to serious identity underreporting problems.

In addition, as time goes by, the age span between the collected registration image and the tested test image changes more and more, and the accuracy of the overall identification becomes lower and lower.

[0004] In addition, pedestrian re-identification ReID in the prior art is a technology similar to face recognition, which mainly relies on human body information for feature extraction and comparison, and is more robust to short-term face occlusion, large angles, etc., but cannot In response to changes such as changing clothes, it is generally used to determine the trajectory of personnel, and the identity information of personnel cannot be determined during the whole process

[0005] It can be seen that there are technical problems of low accuracy in existing identification

Images