3D Object Recognition Using Position and Color Point Cloud Descriptors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object recognition methods face challenges in distinguishing objects with similar shapes or patterns, particularly when objects are inclined or have complex backgrounds, and are inefficient due to reliance on brightness values or edge extraction.

Innovation Solution

A method that extracts feature points by analyzing the position and color information relationships between points in a point cloud, using mathematical formulas to generate descriptors that enhance recognition rates by grouping points and selecting K-nearest neighboring points.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If SIFT or SURF methods are used to extract feature points based on brightness values, then feature points can be extracted, but objects with similar shapes or patterns cannot be distinguished and recognition rate decreases when objects are inclined

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidability to distinguish objects with similar shapes
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent combines multiple feature types (brightness values, color information, position relationships, and shape characteristics) into a unified descriptor. This merging of different feature dimensions allows the system to distinguish objects with similar shapes by utilizing color and position variations, while also maintaining robustness to inclination through the integrated shape information.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent transitions from traditional 2D image space to 3D point cloud space by incorporating depth information. This dimensional change enables the system to capture spatial relationships and object geometry more effectively, improving the ability to distinguish objects with similar appearances but different spatial configurations.

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

2Shape

If edge extraction methods are used to determine object shape, then object shape can be inferred, but object recognition rate decreases when external shape is simple and computational time increases when background is complex

Engineering Contradiction:
Improveobject shape informationVSAvoidobject recognition speed
Core Design Contradiction:
ShapeVSProductivity

Solution Approach 1:

The patent extracts only the most discriminative features from the point cloud data - specifically position relationships and color information - rather than processing all possible shape characteristics. This selective extraction of key features maintains shape information effectiveness while reducing computational complexity and processing time.

Inventive Principle:
Principle #2Taking out (Extraction)

3Shape

If 3D point cloud methods are used to recognize objects based on shape, then three-dimensional object recognition is achieved, but objects with similar shapes cannot be distinguished

Engineering Contradiction:
Improvethree-dimensional shape recognitionVSAvoidobject distinction accuracy
Core Design Contradiction:
ShapeVSMeasurement precision

Solution Approach 1:

The patent creates a composite descriptor that integrates multiple feature types (brightness, color, position, and shape) similar to how composite materials combine different properties. This composite approach allows the system to maintain 3D shape recognition capabilities while adding color and position information to distinguish objects with similar geometries.

Inventive Principle:
Principle #40Composite materials

Data Source

PatentEP2610783B1Object recognition method using an object descriptor
Publication Date: 2020.08.05 SAMSUNG ELECTRONICS CO LTD
  • EP2610783B1 patent drawingFigure 1
  • EP2610783B1 patent drawingFigure 2
  • EP2610783B1 patent drawingFigure 3

AI summary

An object recognition method, a descriptor generating method for object recognition, and a descriptor for object recognition capable of extracting feature points using the position relationship and colour information relationship between points in a group that are sampled from an image of an object, and capable of recognizing the object using the feature points, the object recognition method including extracting feature components of a point cloud using the position information and the colour information of the points that compose the point cloud of the three-dimensional (3D) image of an object, generating a descriptor configured to recognize the object using the extracted feature components; and performing the object recognition based on the descriptor.