Foreground Mask Correction Using Depth and Color Fusion

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional image processing methods for foreground object segmentation in video surveillance and robotics face challenges such as the need for static background generation, which is cumbersome and prone to errors due to noise in depth sensors, leading to incomplete or inaccurate object boundaries.

Innovation Solution

An image-processing device and method that dynamically estimates and corrects foreground masks using depth information, binarizing differences between input frames and background images, and applies a mean shift filter to refine mask values, allowing for real-time segmentation without relying on pre-generated static backgrounds.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a pre-determined static background image is generated by capturing multiple images and taking an average, then the background can be established for segmentation, but the process is cumbersome and requires the user to ensure no moving objects are present during capture

Engineering Contradiction:
Improvebackground generation accuracyVSAvoidbackground generation convenience
Core Design Contradiction:
Measurement precisionVSEase of operation

Solution Approach 1:

The system performs automatic background generation and updating without requiring user intervention. The background image is dynamically updated by the system itself using the most recent frame where the object is not present, eliminating the need for users to manually capture background images or ensure no moving objects are present during capture

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The background image is transformed from a static pre-determined image to a dynamic image that automatically updates over time. The system continuously adapts the background image based on temporal information from multiple frames, allowing the background to evolve and remain accurate without manual regeneration

Inventive Principle:
Principle #15Dynamics

2Device complexity

If a completely static background image is used for segmentation, then the segmentation process is simple, but the background image must be regenerated if the image capture device is displaced, which is cumbersome

Engineering Contradiction:
Improvesegmentation process simplicityVSAvoidcamera displacement adaptability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The background image dynamically adapts to camera displacement automatically. When the capture device moves, the system detects the displacement and updates the background image using temporal information from subsequent frames, maintaining segmentation accuracy without requiring manual background regeneration

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system uses feedback from temporal analysis of multiple frames to detect camera displacement and automatically adjust the background image. By analyzing frame differences and object motion patterns over time, the system determines when background updating is needed and performs the update automatically

Inventive Principle:
Principle #23Feedback

3Productivity

If depth information from a depth sensor is used for segmentation, then object segmentation can be achieved, but the severe noise in depth sensors results in non-smooth boundaries and undesired holes within foreground object regions

Engineering Contradiction:
Improvesegmentation speedVSAvoidobject boundary smoothness
Core Design Contradiction:
ProductivityVSManufacturing precision

Solution Approach 1:

The system introduces color information as an intermediary to bridge the gap between depth-based segmentation and smooth boundary generation. By combining depth information for initial segmentation with color information for boundary refinement, the system achieves both fast segmentation and smooth, accurate object boundaries without the holes and artifacts caused by noisy depth data alone

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The segmentation result is formed by combining multiple information sources (depth information and color information) into a composite segmentation mask. This composite approach leverages the strengths of each modality while compensating for their weaknesses, producing smooth boundaries and complete object regions

Inventive Principle:
Principle #40Composite materials

4Productivity

If conventional background subtraction methods are used, then foreground objects can be segmented, but the methods rely heavily on depth values which contain severe noise, leading to inaccurate boundaries

Engineering Contradiction:
Improvesegmentation efficiencyVSAvoidboundary accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

Color information serves as an intermediary that mediates between the noisy depth-based segmentation and the final accurate boundary. The system uses color data to refine and correct the segmentation mask generated from depth information, eliminating the reliance on noisy depth values for boundary determination while maintaining segmentation efficiency

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The system changes the parameters used for boundary determination from relying heavily on depth values to incorporating color information. By adjusting the weighting and utilization of different data modalities, the system achieves accurate boundaries without sacrificing segmentation speed

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP3327668B1Image-processing device and method for foreground mask correction for object segmentation
Publication Date: 2020.01.08 SONY GROUP CORP
  • EP3327668B1 patent drawingFigure 1
  • EP3327668B1 patent drawingFigure 2
  • EP3327668B1 patent drawingFigure 3

AI summary

An image-processing device and method for foreground mask correction for object segmentation, includes receipt of a sequence of image frames. A first FG mask is estimated using depth information associated with the input image frame and by binarizing a difference between the input image frame and a background (BG) image of the input image frame. A first set of pixels with a first mask value to be updated to a second mask value is identified in a boundary region of the estimated first FG mask. A second FG mask is determined based on the identification of the first set of pixels in the boundary region of the estimated first FG mask.