3D Pose Estimation Using Local Image Patches

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for three-dimensional pose estimation from depth images are hindered by noise and background clutter, leading to inaccurate global image descriptors and poor representation of object poses.

Innovation Solution

The approach involves generating local feature databases for each image patch, rather than a single global feature database, to reduce the influence of noise and background clutter. Local image patches are sampled from depth images, and patch-based features are extracted using a convolutional neural network. These features are then searched in corresponding local feature databases to identify nearest neighbor poses, which are used to estimate the three-dimensional pose of the object.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If global image descriptors are used for pose estimation, then the method is simple to implement, but the accuracy deteriorates due to noise and background clutter

Engineering Contradiction:
Improveease of implementationVSAvoidpose estimation accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The patent divides the depth image into multiple local patches and creates separate feature databases for each patch type (e.g., corner patches, edge patches, center patches). This segmentation allows the system to focus on local features that are less affected by global noise and background clutter, thereby improving pose estimation accuracy while maintaining implementation feasibility through systematic processing.

Inventive Principle:
Principle #1Segmentation

2Measurement precision

If local image patches are used for pose estimation, then the accuracy improves by reducing noise influence, but the device complexity increases

Engineering Contradiction:
Improvepose estimation accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies different processing strategies to different types of local patches based on their specific characteristics. For example, corner patches may use different feature extraction parameters compared to center patches. This local quality approach improves accuracy by optimizing each patch type for its specific role while managing overall system complexity through targeted rather than universal processing.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent creates a unified framework that handles multiple patch types using a consistent overall process flow. The same convolutional neural network architecture processes all patches, and the same nearest-neighbor search methodology is applied across all feature databases. This universality reduces implementation complexity despite the presence of multiple local feature databases.

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

3Measurement precision

If multiple local feature databases are created for different patch types, then the representation precision improves, but the loss of time increases due to multiple searches

Engineering Contradiction:
Improvepose representation precisionVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent pre-processes training images to extract local patches and build separate feature databases for each patch type before actual pose estimation is needed. This preliminary action organizes the data structure in advance, so that during runtime, the system only needs to perform nearest-neighbor searches in pre-organized databases rather than processing raw images, significantly reducing processing time while maintaining high precision.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3586311B1Image indexing and retrieval using local image patches for object three-dimensional pose estimation
Publication Date: 2025.04.02 SIEMENS MOBILITY GMBH
  • EP3586311B1 patent drawingFigure 1
  • EP3586311B1 patent drawingFigure 2
  • EP3586311B1 patent drawingFigure 3

AI summary

The present embodiments relate to automatically estimating a three]dimensional pose of an object from an image captured using a camera with a structured light sensor. By way of introduction, the present embodiments described below include apparatuses and methods for training a system for and estimating a pose of an object from a test image. Training and test images are sampled to generate local image patches. Features are extracted from the local image patches to generate feature databased used to estimate nearest neighbor poses for each local image patch. The closest nearest neighbor pose to the test image is selected as the estimated three]dimensional pose.