Video Behavior Prediction Using Object Relationships and Facial Cues

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing behavior recognition technologies are limited to recognizing behaviors that have already occurred, making it difficult to take timely countermeasures, such as preventing shoplifting or detecting early signs of illness, as they only provide post-detection services.

Innovation Solution

An information processing apparatus that combines behavior analysis with context sensing to predict future behaviors by analyzing current facial expressions and relationships with the environment and objects, using machine learning models like hidden Markov models to anticipate potential events.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If behavior recognition technology is used to recognize behaviors from video image data, then behavior detection capability is improved, but the ability to detect situations requiring countermeasures in advance deteriorates

Engineering Contradiction:
Improvebehavior detection capabilityVSAvoidtime for countermeasure
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system performs preliminary action by predicting future behaviors before they actually occur. The behavior prediction unit uses the relationship model and probability model to anticipate what a person will do next based on their current behavior and contextual relationships, enabling countermeasures to be prepared in advance rather than reacting after the behavior occurs.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The system applies beforehand cushioning by preparing countermeasures in advance based on predicted behaviors. When the behavior prediction unit forecasts a potential problematic behavior, the system can pre-activate appropriate countermeasures or alerts, cushioning against the potential negative outcome before it materializes.

Inventive Principle:
Principle #11Beforehand cushioning (Prior cushioning)

2Measurement precision

If traditional behavior recognition is used, then current behavior detection is achieved, but future behavior prediction capability is lost

Engineering Contradiction:
Improvecurrent behavior detection accuracyVSAvoidfuture behavior information
Core Design Contradiction:
Measurement precisionVSLoss of information

Solution Approach 1:

The system uses preliminary action by predicting future behaviors based on current observations. The probability model calculates the likelihood of various future behaviors occurring, allowing the system to infer information about future actions from current behavior and relationship data before those actions actually take place.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP4207097B1Information processing program, information processing method, and information processing apparatus
Publication Date: 2025.10.22 FUJITSU LTD
  • EP4207097B1 patent drawingFigure 1
  • EP4207097B1 patent drawingFigure 2
  • EP4207097B1 patent drawingFigure 3

AI summary

An information processing apparatus acquires (51) video image data that includes target objects including a person and an object, first specifies (52), by inputting the acquired video image data to a first machine learning model, a relationship between each of the target objects included in the acquired video image data, second specifies (53), by using a feature value of the person included in the acquired video image data, a behavior of the person included in the video image data, and predicts (54), by inputting the specified behavior of the person and the specified relationship to a probability model, a future behavior or a future state of the person.