Depth-Color Fusion for Target Object Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image processing technologies for extracting a target object area, such as a human body, in virtual reality applications are limited by the need for expensive and non-portable 3D scanning equipment, and image segmentation using only depth or color images lacks precision.
Innovation Solution
An image processing apparatus and method that extracts a silhouette area from a depth image and refines it using a color image, involving a floor removal unit, depth filtering, silhouette extraction, and probability calculation to accurately identify and extract the target object area, including a post-processor for refinement and image matching when necessary.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D scanning equipment is used to extract target object area, then measurement precision is improved, but device complexity and cost increase
Solution Approach 1:
The patent combines depth image and color image processing into a unified system. The depth image provides spatial structure information while the color image provides surface appearance information, and their fusion enables accurate target object area extraction without requiring complex 3D scanning equipment. The merging of these two image types allows the system to achieve measurement precision previously only attainable with expensive 3D scanners.
Solution Approach 2:
The patent introduces an intermediary processing stage that integrates depth and color image information. By using the depth image to guide color image segmentation and vice versa, the system creates a mediator mechanism that enables accurate target extraction through combined information processing, avoiding the need for direct complex 3D scanning hardware.
2Device complexity
If only depth image is used for image segmentation, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent merges depth image processing and color image processing into a unified segmentation system. The depth image provides structural information for initial segmentation, while the color image provides surface information for refinement. This combination achieves higher measurement precision than depth-only methods while maintaining relatively simple device complexity by processing standard 2D images.
Solution Approach 2:
The patent applies segmentation by dividing the extraction process into multiple stages: initial segmentation based on depth image, followed by refinement based on color image. This multi-stage segmentation approach allows each image type to contribute its strengths while maintaining overall system simplicity and achieving high precision results.
3Device complexity
If only color image is used for image segmentation, then device complexity is reduced, but measurement precision deteriorates
Solution Approach 1:
The patent combines color image processing with depth image processing to overcome the limitations of color-only segmentation. The depth image provides spatial boundaries and structure information that complements color information, enabling more precise target object area extraction while maintaining simple device architecture through standard image processing techniques.
Solution Approach 2:
The patent performs preliminary segmentation using the depth image before refining with the color image. This preliminary action establishes the structural framework and spatial boundaries, which then guide the color-based refinement process. This sequence allows each image type to be utilized effectively in its optimal role, achieving high precision without complex equipment.
Data Source
AI summary
Provided is an image processing apparatus, method and computer-readable medium. The image processing apparatus, method and computer-readable medium may extract a target object area from an input color image, based on an input depth image and the input color image. For the above image processing, the image processing may extract a silhouette area of a target object from the input depth image and refine the silhouette area of the target object based on the input color image.