Virtual Fitting via Facial Posture Tracking and Affine Transformation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current virtual fitting techniques either lack accuracy without actual body information or require specialized devices for capturing human body data, limiting their accessibility and usability for general users.

Innovation Solution

A method and device utilizing a common image capturing device, such as a camera, to detect facial postures and adjust article images (like glasses or jewelry) to match the user's facial posture, allowing for virtual fitting by tracking feature points and performing affine transformations to maintain consistency, enabling users to view wearing effects at multiple angles.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of operation

If a synthetic model is used for virtual fitting, then the fitting process can be implemented without actual body information, but the fitting effect is poor and lacks accuracy

Engineering Contradiction:
Improveease of operationVSAvoidfitting accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent uses a camera to capture the user's actual facial image and creates a digital copy of the user's face. This copied facial model is then used for virtual fitting, combining the ease of using common devices with the accuracy of actual body information. The system extracts facial features from the captured image to build an accurate 3D facial model for rendering accessories.

Inventive Principle:
Principle #26Copying

2Measurement precision

If a special device such as a depth of field sensor is used to capture actual body information, then the fitting accuracy is improved, but the device complexity increases and requires specialized equipment not possessed by general users

Engineering Contradiction:
Improvebody information accuracyVSAvoiddevice complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent makes the virtual fitting system universally accessible by using a common camera device that most users already possess. The camera serves multiple functions: capturing facial images, extracting feature points, and providing sufficient data for accurate 3D facial modeling. This eliminates the need for specialized depth sensors while maintaining fitting accuracy through intelligent image processing and feature extraction algorithms.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Reliability

If feature point tracking and affine transformation are used to maintain article-image consistency with facial posture, then the virtual fitting realism is improved, but the computational complexity increases

Engineering Contradiction:
Improvevirtual fitting realismVSAvoidcomputational power
Core Design Contradiction:
ReliabilityVSPower

Solution Approach 1:

The patent segments the facial model into key feature points (landmarks) that define the facial structure and posture. By tracking only these critical feature points rather than processing the entire facial surface, the system maintains computational efficiency while achieving accurate article-to-face alignment. The segmentation approach allows for real-time posture tracking and affine transformation with reduced computational burden.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentUS10360731B2Method and device for implementing virtual fitting
Publication Date: 2019.07.23 BEIJING JINGDONG SHANGKE INFORMATION TECH CO LTD
  • US10360731B2 patent drawing
  • US10360731B2 patent drawing
  • US10360731B2 patent drawing

AI summary

The invention provides methods and devices for implementing virtual fitting utilizing an image capturing device such as a camera on a mobile phone or a computer. The method for implementing virtual fitting of the invention comprises: detecting a face on a captured initial frame, generating an article image at an initial position when the face has been detected, and then superimposing the article image on the initial frame for output, the initial position superimposed on a specific position of the face in the initial frame; detecting a facial posture on the face in the current frame to obtain the facial posture of the current frame; and generating the article image again according to a current position of the article image and the facial posture, making an article posture in the article image be consistent with the facial posture, and then superimposing the article image on the current frame for output.