Automated Offensive Action Detection in Visual Sessions

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing visual communication systems face challenges in consistently detecting and addressing offensive actions within mixed reality, augmented reality, and virtual reality sessions, as subjective determinations can lead to inconsistent enforcement and negative stigma against reporting objectionable behavior, resulting in unresolved offensive actions that affect other users.

Innovation Solution

A processing system that passively observes participants' activities to detect objectionable actions using machine learning algorithms and action detection models, which can identify patterns in visual features and wearable device inputs, allowing for automated filtering and modification of visual content to block or modify offensive actions without manual annotation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual annotation and human review are used to detect offensive actions, then measurement precision can be maintained, but productivity decreases and loss of time increases

Engineering Contradiction:

Improvedetection accuracy of offensive actionsVSAvoidspeed of detecting and addressing offensive actions

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent replaces manual human review processes with automated machine learning-based detection systems. The processing system uses action detection models that analyze visual content from communication sessions to automatically identify offensive actions, eliminating the need for manual annotation while maintaining detection accuracy and significantly improving processing speed.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system performs self-service by automatically detecting, analyzing, and addressing offensive actions without requiring human intervention. The action detection models continuously monitor visual content, identify patterns associated with offensive behavior, and execute modifications autonomously based on predefined policies and user configurations.

Inventive Principle:

Principle #25Self-service

2Productivity

If automated detection systems are implemented, then productivity increases, but device complexity increases

Engineering Contradiction:

Improveautomated detection and filtering of offensive actionsVSAvoidcomplexity of processing system

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the detection system into multiple independent action detection models, each specialized in identifying specific types of offensive actions. This modular architecture allows the system to handle complex detection tasks by combining simpler, specialized models rather than using a single complex monolithic system.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The processing system is designed as a universal platform that can detect multiple types of offensive actions through different action detection models. The same core system infrastructure supports various detection functions by loading different models, reducing overall system complexity compared to having separate specialized systems for each detection task.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Reliability

If human reviewers investigate complaints, then reliability can be maintained through subjective judgment, but loss of time increases and productivity decreases

Engineering Contradiction:

Improveconsistency in determining offensive actionsVSAvoidtime for investigating and resolving complaints

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The system incorporates feedback mechanisms where users can report offensive actions, and the action detection models use this feedback to refine their detection accuracy. The processed visual content and detection results provide continuous feedback loops that improve the system's reliability over time while maintaining rapid processing speeds.

Inventive Principle:

Principle #23Feedback

Solution Approach 2:

The action detection models perform preliminary analysis of visual content in real-time during communication sessions, identifying potential offensive actions before they escalate or affect other users. This proactive detection eliminates the need for post-event human investigation, reducing time loss while maintaining consistent determination through algorithmic analysis.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS11341775B2Identifying and addressing offensive actions in visual communication sessions

Publication Date: 2022.05.24 AT&T INTELLECTUAL PROPERTY I L P

AI summary

A processing system having at least one processor may establish a communication session between a first communication system of a first user and a second communication system of a second user, the communication session including first visual content, the first visual content including a first visual representation of the first user, and detecting a first action of the first visual representation in the first visual content in accordance with a first action detection model. The processing system may modify, in response to the detecting the first action, the first visual content in accordance with a first configuration setting of the first user for the communication session, which may include modifying the first action of the first visual representation of the first user in the first visual content. In addition, the processing system may transmit the first visual content that is modified to the second communication system of the second user.