Pedestrian re-identification method and system based on depth conditional random field

Abstract

The invention discloses a pedestrian re-identification method based on a depth conditional random field, and the method comprises the steps: obtaining a pedestrian re-identification data set, obtaining all pedestrian pictures corresponding to each pedestrian in K pedestrians from the pedestrian re-identification data set, and obtaining a pedestrian picture corresponding to each pedestrian in the Kpedestrians for each pedestrian in the K pedestrians; randomly obtaining one pedestrian picture from the corresponding pedestrian pictures as a target pedestrian picture, and obtaining N pedestrian pictures except the target pedestrian picture from the corresponding pedestrian pictures to form a positive sample picture set corresponding to the pedestrian, m pedestrian pictures not corresponding to the pedestrian are acquired from the pedestrian re-identification data set to form a negative sample picture set corresponding to the pedestrian, and all 1 + M + N pedestrian pictures jointly form apicture group corresponding to the pedestrian. The pedestrian re-identification method can solve the technical problems of low pedestrian re-identification accuracy and difficulty in solving parameters of the graph model in the existing pedestrian re-identification method of the graph model.

Description

technical field [0001] The invention belongs to the field of pattern recognition, and more specifically relates to a pedestrian re-identification method and system based on a depth conditional random field. Background technique [0002] Person Re-Identification (ReID) is a popular research topic in the field of computer vision, which aims to retrieve target pedestrians from different cameras in overlapping areas without viewing angles. Therefore, in video surveillance, social security, etc. play an important role. [0003] Traditional person re-identification methods are mainly divided into two stages in turn, namely feature extraction stage and distance measurement stage. Among them, in the first stage, the cropped pedestrian pictures are sent to the network model to extract feature vectors, which often have certain robustness and distinguishability, and can roughly distinguish different pedestrian identities; In the second stage, the above-mentioned feature vectors of pe...

AI Technical Summary

Problems solved by technology

Its purpose is to transform the traditional pedestrian re-identification problem based on the distance measurement between two pictures into the category state labeling problem under the CRF model through the construction of a graph model, so as to mine the correlation between data to obtain a global optimal solution. Thereby solving the technical problem of pedestrian re-identification accuracy existing in the traditional pedestrian re-identification method; at the same time, the present invention embeds the potential function and state reasoning of CRF into the neural network in the process of solving the CRF vertex category, and utilizes the powerful learning ability of the neural network and adaptive optimization mechanism to solve the technical problems of low pedestrian re-identification accuracy and difficulty in solving graphical model parameters existing in existing graphical model pedestrian re-identification methods

Images