3D Feature Point Extraction Using Depth Image Geometry

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional feature point extraction and tracking technologies using brightness images suffer from low repeatability due to varying illumination conditions and camera viewpoints, making it difficult to match features between images or track points between video frames.

Innovation Solution

An image processing apparatus that analyzes geometry information from depth images to extract 3D feature points by calculating spatial coordinates and surface normal vectors, comparing these vectors to identify feature point candidates, and selecting points with significant angle differences as feature points, thereby enhancing repeatability and enabling effective matching and tracking across varying conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If feature point extraction is performed using brightness images, then the extraction process can be implemented with conventional technologies, but the repeatability of feature point extraction becomes low due to varying illumination conditions

Engineering Contradiction:

Improverepeatability of feature point extractionVSAvoidillumination condition variations

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The patent changes the parameter basis for feature point extraction from brightness/intensity values to geometric parameters (spatial coordinates and surface normal vectors) derived from depth images. This parameter transformation makes the feature points invariant to illumination changes, thereby improving repeatability while eliminating sensitivity to lighting conditions

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent replaces the optical-based brightness image acquisition system with a depth image acquisition system that measures geometric properties directly. By substituting optical intensity measurement with geometric measurement, the system achieves illumination-invariant feature extraction

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If feature point extraction is performed using brightness images, then the existing image processing pipeline can be maintained, but the matching accuracy between images with different viewpoints deteriorates

Engineering Contradiction:

Improvefeature point matching accuracyVSAvoidfeature point consistency across viewpoints

Core Design Contradiction:

Measurement precisionVSStability of the object's composition

Solution Approach 1:

The patent transforms feature representation from 2D brightness-based descriptors to 3D geometric descriptors (spatial coordinates and surface normal vectors). These geometric parameters remain stable across different viewpoints and provide more reliable matching accuracy compared to brightness-based features that vary with viewing angle

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The patent transitions from 2D brightness image analysis to 3D depth image analysis by incorporating the depth dimension. This dimensional expansion provides additional geometric information (surface normal vectors) that enhances feature point matching accuracy and viewpoint invariance

Inventive Principle:

Principle #17Another dimension (Dimensionality change)

Data Source

PatentUS9165211B2Image processing apparatus and method

Publication Date: 2015.10.20 SAMSUNG ELECTRONICS CO LTD

AI summary

Provided is an image processing apparatus for extracting a three-dimensional (3D) feature point from a depth image. An input processing unit may receive a depth image and may receive, via a user interface, selection information of at least one region that is selected as a target region in the depth image. A geometry information analyzer of the image processing apparatus may analyze geometry information of the target region within the input depth image, and a feature point extractor may extract at least one feature point from the target region based on the geometry information of the target region.