Image Feature Estimation Using Matrix Transformations and Convolutional Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for estimating image characteristics, such as human posture, in computer vision technologies face challenges with low accuracy due to high hardware requirements and high error rates, especially in dynamic environments.

Innovation Solution

The method involves extracting eigenvalues from input image data, including matching degrees, probability of characteristic co-occurrence, and distance scores, followed by multiple matrix transformations and convolutional network calculations to improve estimation accuracy using fully-connected belief networks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a tailored shooting apparatus with depth information is used to capture image data, then estimation accuracy of image characteristics is improved, but hardware performance requirements and device complexity increase

Engineering Contradiction:
Improveestimation accuracyVSAvoidhardware performance requirements
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts depth information from ordinary 2D images through computational methods rather than requiring specialized depth-capturing hardware. By using image processing algorithms to derive depth maps from monocular images, the system achieves 3D-aware pose estimation without tailored shooting apparatus, thus improving estimation accuracy while reducing hardware complexity requirements

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent replaces mechanical/optical depth-capturing systems with computational image processing methods. Instead of using structured light, stereo cameras, or other hardware-based depth acquisition mechanisms, the system substitutes these with algorithmic approaches that analyze 2D image data to infer 3D characteristics, thereby eliminating the need for complex hardware while maintaining estimation accuracy

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If a tailored shooting apparatus with specific distance requirements is used, then estimation accuracy is improved, but adaptability to various practical needs deteriorates

Engineering Contradiction:
Improveestimation accuracyVSAvoidflexibility in practical application
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent creates a universal pose estimation system that can handle various shooting distances and scenarios using the same algorithmic framework. The method adapts to different practical needs by processing ordinary images regardless of capture conditions, making the system versatile for security monitoring, medical rehabilitation, sports analysis, and other applications without requiring distance-specific hardware configurations

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

Solution Approach 2:

The patent employs parameter adaptation within the algorithm to accommodate different shooting distances and scenarios. By dynamically adjusting computational parameters and using scale-invariant feature detection, the system maintains estimation accuracy across varying distances without imposing fixed distance requirements, thereby achieving both precision and adaptability

Inventive Principle:
Principle #35Parameter changes

3Ease of manufacture

If static image analysis methods such as edge detection and image histogram are used, then processing simplicity is maintained, but error rate in real-time dynamic image data increases

Engineering Contradiction:
Improveprocessing simplicityVSAvoiderror rate in characteristic estimation
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The patent transitions from static image analysis to dynamic pose estimation by incorporating temporal information and motion cues. The system processes video sequences and uses motion consistency across frames to improve estimation reliability in dynamic environments, while maintaining computational efficiency through optimized algorithms that handle real-time data without excessive complexity

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent employs continuous pose estimation across video frames rather than analyzing isolated static images. By maintaining temporal coherence and using prediction algorithms that leverage information from previous frames, the system reduces errors in dynamic scenarios while keeping processing requirements manageable through efficient recurrent computation strategies

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP3121788B1Image feature estimation method and device
Publication Date: 2020.11.04 HUAWEI TECH CO LTD
  • EP3121788B1 patent drawingFigure 1
  • EP3121788B1 patent drawingFigure 2~3
  • EP3121788B1 patent drawing

AI summary

The present invention discloses an image characteristic estimation method and device, where content of the method includes: extracting at least two eigenvalues of input image data; executing the following operations for each extracted eigenvalue, until execution for the extracted eigenvalues is completed: selecting an eigenvalue, and performing at least two matrix transformations on the eigenvalue by using a pre-obtained matrix parameter, to obtain a first matrix vector corresponding to the eigenvalue; when a first matrix vector corresponding to each extracted eigenvalue is obtained, obtaining second matrix vectors with respect to the at least two extracted eigenvalues by using a convolutional network calculation method according to the obtained first matrix vector corresponding to each eigenvalue; and obtaining a status of an image characteristic in the image data by means of estimation according to the second matrix vectors. In this way, accuracy of estimation is effectively improved.