首页>
外国专利>
PERSON RE-IDENTIFICATION METHOD BASED ON CONSISTENCY CONSTRAINT FEATURE LEARNING
PERSON RE-IDENTIFICATION METHOD BASED ON CONSISTENCY CONSTRAINT FEATURE LEARNING
展开▼
机译:基于一致性约束特征学习的人员重新识别方法
展开▼
页面导航
摘要
著录项
相似文献
摘要
A person re-identification method based on consistency constraint feature learning, comprising: acquiring a picture of a person by means of a camera network and marking a training set, and setting a parameter and initializing a convolutional neural network, wherein the camera network constructs a plurality of camera pairs (S1); sampling a subset of pictures from a database, extracting feature information by using the convolutional neural network, and calculating and obtaining a similarity matrix of all people according to the feature information (S2); solving the optimal matching of a relationship matrix of all the people according to a pre-set objective function and a pre-set gradient descent method; obtaining gradient back propagation according to a deviation between the optimal matching of the relationship matrix of all the people and a relationship matrix, obtained according to actual situations, of all the people, so as to train the convolutional neural network according to the gradient back propagation (S4); and repeating steps S2-S4 until the user requirements are met (S5). The method can adapt to an application scenario of matching under a large camera network, thereby eliminating a contradictory matching error.
展开▼