Human Body Attribute Recognition via Heat Map Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current human body attribute recognition technologies often result in inaccurate results due to the misidentification of body features, such as the head being mistaken for the torso, leading to incorrect attribute recognition.

Innovation Solution

A method and apparatus that utilize a recognition model to generate heat maps, fuse global and local heat maps, and correct the focus area for each attribute, improving the accuracy of human body attribute recognition by ensuring each attribute focuses on the appropriate area.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If recognition is performed on multiple human body features simultaneously, then comprehensive attribute recognition is achieved, but misidentification between connected body parts occurs

Engineering Contradiction:
Improvecomprehensive attribute recognitionVSAvoidbody feature identification accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent divides the human body into multiple independent feature regions (head, torso, limbs) and processes each region separately through dedicated recognition branches. This segmentation prevents misidentification between connected body parts while maintaining comprehensive attribute recognition capability across all body regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies different recognition strategies and attention mechanisms to different body regions based on their specific characteristics. Each body part receives customized processing tailored to its local features, improving identification accuracy while maintaining overall system versatility.

Inventive Principle:
Principle #3Local quality

2Adaptability or versatility

If the recognition model focuses on multiple attributes simultaneously, then comprehensive attribute recognition is achieved, but the focus area becomes diffuse and less accurate

Engineering Contradiction:
Improvemulti-attribute recognition capabilityVSAvoidattribute focus accuracy
Core Design Contradiction:
Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent segments the recognition process into multiple independent branches, each dedicated to specific body regions or attribute types. This allows the model to maintain sharp focus on individual attributes while preserving the ability to recognize multiple attributes comprehensively through the integrated multi-branch architecture.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces attention mechanisms and heat map visualization that add a spatial dimension to attribute recognition. By mapping attribute importance across different body regions, the model can simultaneously handle multiple attributes while maintaining precise focus through dimensional expansion of the recognition space.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP3989112B1Human body attribute recognition method and apparatus, electronic device and storage medium
Publication Date: 2025.01.15 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3989112B1 patent drawingFigure 1A~1B
  • EP3989112B1 patent drawingFigure 2A~2B
  • EP3989112B1 patent drawingFigure 2C~2D

AI summary

Disclosed in the embodiments of the present application are a human body attribute recognition method and apparatus, an electronic device and a storage medium. The human body attribute recognition method is executed by the electronic device, and comprises: acquiring a human body image sample comprising a plurality of detection regions, the detection regions being labeled with real values of human body attributes; generating a thermal image of the human body image sample and a thermal image of a region to be detected by means of a recognition model so as to obtain a global thermal image and a local thermal image corresponding to the human body image sample; fusing the global thermal image and the local thermal image, and performing human body attribute recognition on the fused image to obtain predicted values of the human body attributes of the human body image sample; determining a region of interest for each type of human body attribute according to the global thermal image and the local thermal image; correcting the recognition model by using the regions of interest, the real values of the human body attributes and the predicted values of the human body attributes; and on the basis of the corrected recognition model, performing human body attribute recognition on an image to be recognized.