Pose Identification Using Coordinate Down-Sampling

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current pose identification methods using depth images face limitations in precision due to the inability to effectively utilize depth information and require significant memory and computational resources, leading to low-resolution feature maps and decreased accuracy.

Innovation Solution

The proposed method involves obtaining 3D coordinate information corresponding to features in depth images, using a coordinate maintenance module to perform feature and coordinate down-sampling simultaneously, and integrating this information into the pose identification process to enhance precision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional feature extraction methods are used on depth images, then computational resources and memory are significantly consumed, but the output feature maps become low-resolution leading to decreased pose identification accuracy

Engineering Contradiction:
Improvepose identification accuracyVSAvoidcomputational resource consumption
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the feature extraction process into multiple resolution levels. A down-sampling operation is performed to generate a lower-resolution feature map that requires fewer computational resources, while position information is preserved and integrated to maintain accuracy. This segmentation allows the system to process features at different scales efficiently.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces position information as an additional dimension to the feature extraction process. By incorporating spatial position data alongside feature values, the system can maintain accurate pose identification even when using down-sampled lower-resolution feature maps, thus resolving the trade-off between computational efficiency and accuracy.

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

2Productivity

If traditional feature extraction methods are used on depth images, then computational resources and memory are significantly consumed, but the pose identification accuracy decreases due to low-resolution feature maps

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidpose identification accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The feature extraction process is divided into stages with down-sampling operations that reduce computational load while preserving essential information. The patent processes features at multiple resolution levels rather than maintaining full resolution throughout, improving processing efficiency.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Position information is added as a critical dimension to the feature representation. This allows the system to achieve accurate pose identification using down-sampled feature maps by incorporating spatial location data, thus maintaining precision while improving productivity.

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

3Device complexity

If feature down-sampling is performed to reduce computational requirements, then memory and processing resources are reduced, but position information of features is lost

Engineering Contradiction:
Improvememory requirementsVSAvoidposition information
Core Design Contradiction:
Device complexityVSLoss of information

Solution Approach 1:

The patent performs down-sampling operations in a controlled manner while preserving position information. By预先 (in advance) planning the down-sampling strategy to maintain spatial coordinates, the system reduces memory requirements without losing critical position data needed for accurate pose identification.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

Position information is maintained as a separate dimensional component during down-sampling. Rather than losing spatial data when reducing feature map resolution, the patent preserves position information as an additional dimension that can be integrated with the down-sampled features.

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

Data Source

PatentEP3965071B1Method and apparatus for pose identification
Publication Date: 2025.01.15 SAMSUNG ELECTRONICS CO LTD
  • EP3965071B1 patent drawingFigure 1
  • EP3965071B1 patent drawingFigure 2A
  • EP3965071B1 patent drawingFigure 2B

AI summary

Disclosed is a pose identification method including obtaining a depth image of a target, obtaining feature information of the depth image and position information corresponding to the feature information, and obtaining a pose identification result of the target based on the feature information and the position information.