Clothing Analysis Method and System for Specific Pedestrians Based on Fashion Graph Migration

Abstract

The invention discloses a specific pedestrian clothing analysis method and system based on fashion map migration, including: first, the model is initialized by training the fashion data set; then, the model is used to analyze monitoring images containing pedestrians, and the strip Constrained optimization parses the results; finally, the model transfer problem is handled by an example-based transfer learning method using weak labels in the monitoring dataset. The method proposed in the present invention can obtain the clothing information in the fashion atlas, and apply it to the monitoring field through transfer learning combined with weak supervision, which can not only retain the information obtained by the clothing analysis in the fashion field, but also improve the analysis effect of the monitoring clothing, and Greatly reduce the workload of data labeling. In the quantitative and qualitative tests on the actual monitoring data set, the validity of the proposed method of the present invention is proved.

Description

technical field [0001] The present invention relates to the technical field of monitoring image segmentation, relates to a pedestrian clothing analysis method, in particular to a specific pedestrian clothing analysis method and system based on fashion map migration. Background technique [0002] Clothing parsing is an image segmentation task for specific clothing and body parts. Although clothing analysis is a research hotspot that has only emerged in recent years in computer vision, it has been widely used in areas such as person recognition, pose estimation, and fashion graph analysis. In particular, clothing analysis related to fashion graphs has achieved relatively successful research results and applications, such as: clothing retrieval, e-commerce, etc. At the same time, although the analysis of the clothes of pedestrians in the surveillance video can obtain key information in the intelligent security system such as pedestrian identity, location, and occlusion, which ...

AI Technical Summary

Problems solved by technology

[0007] Chinese Patent Document No. CN105760999A, published (announcement) date 2016.07.13, discloses a clothing recommendation management method and system. In the clothing extraction stage, it uses the face to locate the position of the human body, and then combines the edge information of the clothing to determine the location of the clothing area. method, and then add clothing labels to the area through clothing recognition, but this method only gets the rectangular area of ​​the clothes, not the pixel-level segmentation area

However, this invention is mainly aimed at the segmentation of the human body, and lacks the mining of clothing information

[0009] Chinese Patent Document No. CN104794722A, open (announcement) date 2015.07.22, discloses the method of using a single Kinect to calculate the three-dimensional net body model of the dressed human body. The invention uses the Kinect camera to capture the RGB-D image of the dressed human body and Skeleton information, denoising after extracting the depth image of the human body, using the depth information, and then dividing the human body into multiple rigid parts, the difference between this method and the present invention is that additional depth information is used, and the category information of clothing cannot be obtained

However, this method is mainly aimed at the analysis of human body parts, and has not further explored specific clothing information.

Images