Salient Object Extraction in Virtual Environments for Low-Bandwidth Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current automatic detection of content in virtual environments is burdensome on computer processing and network bandwidth due to the retrieval and analysis of raw data streams, lacking indication of specific properties to be detected, and raises security and anonymization concerns.

Innovation Solution

A method for automatic extraction of salient objects in virtual environments by determining spatial, visual, and audio properties relative to a reference avatar, using normalized saliency factors to modify or store objects based on a threshold saliency measure, reducing data volume and processing needs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If raw data streams describing all objects in virtual environment are retrieved and analyzed, then detection of particular content can be performed, but computer processing burden and network bandwidth consumption increase significantly

Engineering Contradiction:
Improvedetection accuracyVSAvoidprocessing burden
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts only the relevant properties of salient objects rather than processing all raw data streams. The system identifies and extracts specific properties (visual, spatial, audio) of objects that meet saliency criteria, filtering out unnecessary data before transmission and analysis, thereby reducing processing burden while maintaining detection accuracy

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different processing quality levels to different objects based on their saliency. Salient objects receive detailed property extraction and transmission, while non-salient objects are either omitted or processed with minimal resources. This local differentiation optimizes the balance between detection precision and processing complexity

Inventive Principle:
Principle #3Local quality

2Measurement precision

If all objects and environmental features are analyzed in real-time, then comprehensive detection is achieved, but processing speed and real-time performance deteriorate

Engineering Contradiction:
Improvedetection comprehensivenessVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent performs preliminary saliency assessment and property extraction before full analysis. By pre-identifying salient objects and their relevant properties, the system prepares filtered data in advance, enabling faster real-time processing and comprehensive detection without the computational burden of analyzing all objects simultaneously

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent segments the virtual environment processing into distinct stages: saliency assessment, property extraction, filtering, and transmission. This segmentation allows each stage to operate independently and efficiently, improving overall processing speed while maintaining comprehensive detection capabilities

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If complex multimodal data from different formats is retrieved for machine learning, then detection capability is enhanced, but data bandwidth and storage requirements increase

Engineering Contradiction:
Improvedetection capabilityVSAvoiddata volume
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the essential multimodal properties (visual, spatial, audio) of salient objects rather than transmitting complete raw data streams. By selecting and extracting only the relevant features needed for machine learning detection, the system reduces data volume significantly while preserving detection capability

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies partial action by transmitting only the necessary portion of object data (relevant properties of salient objects) rather than complete datasets. This selective transmission provides sufficient information for effective machine learning detection while minimizing bandwidth and storage requirements

Inventive Principle:
Principle #16Partial or excessive action

4Measurement precision

If detailed properties of all objects are transmitted, then analysis accuracy is improved, but network bandwidth consumption increases

Engineering Contradiction:
Improveanalysis accuracyVSAvoidbandwidth consumption
Core Design Contradiction:
Measurement precisionVSLoss of energy

Solution Approach 1:

The patent extracts and transmits only the relevant properties of salient objects rather than detailed information about all objects. This selective extraction maintains analysis accuracy for important objects while significantly reducing the total data transmitted over the network

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent applies different transmission quality levels to different objects based on saliency. Salient objects receive detailed property transmission with high analysis accuracy, while non-salient objects are either omitted or transmitted with minimal properties, optimizing the balance between analysis accuracy and bandwidth consumption

Inventive Principle:
Principle #3Local quality

Data Source

PatentUS12586341B2Automatic extraction of salient objects in virtual environments for object modification and transmission
Publication Date: 2026.03.24 ROBLOX CORP
  • US12586341B2 patent drawing
  • US12586341B2 patent drawing
  • US12586341B2 patent drawing

AI summary

Automatic extraction of salient object properties in virtual environments for object modification and transmission. In some implementations, a computer-implemented method includes determining a reference avatar and obtaining properties of an object in the virtual environment, the properties including spatial, visual, and/or audio properties. Saliency factors of the object are determined, each saliency factor normalized to a numeric range and based on a different set of properties of the object, where one or more saliency factors are additionally based on a property of the reference avatar. A saliency measure of the object is determined with respect to the reference avatar based on a combination of the saliency factors. If the saliency measure is greater than a threshold saliency measure, the reference avatar or object are automatically modified in the virtual environment based on the object, and otherwise the modification is omitted.