Video Analysis for Abnormal Behavior Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current video monitoring technologies fail to accurately determine abnormal behavior due to limitations in analyzing depth relationships between objects and persons in two-dimensional space, leading to incorrect assessments of individuals performing dangerous actions.

Innovation Solution

An information processing program and device that analyze video frames to accurately identify interactions between objects and persons by using a machine learning model to specify regions and relationships, generating attention maps to contextualize object and person interactions, and determining abnormal behavior based on these relationships.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If two-dimensional Bbox positional relationship analysis is used, then the system is simple and fast, but depth analysis cannot be performed leading to incorrect abnormal behavior detection

Engineering Contradiction:
Improveabnormal behavior detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transitions from two-dimensional Bbox analysis to three-dimensional spatial analysis by introducing depth information through camera calibration parameters and 3D coordinate transformations. This allows the system to determine whether a person is actually on top of an object or merely positioned behind it in the 2D image plane, thereby resolving the false positive problem while maintaining computational efficiency through pre-calibrated camera parameters.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Productivity

If machine learning models are used for object detection, then detection speed is improved, but false positives occur due to inability to distinguish depth relationships

Engineering Contradiction:
Improvedetection speedVSAvoiddetection accuracy
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent introduces an intermediary processing layer that takes the 2D Bbox coordinates from the machine learning detection model and transforms them into 3D spatial relationships using camera calibration data. This intermediary step acts as a bridge between fast 2D detection and accurate 3D reasoning, maintaining the speed advantage of ML models while adding the depth discrimination capability needed to eliminate false positives.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4390874A1Information processing program, information processing method, and information processing device
Publication Date: 2024.06.26 FUJITSU LTD
  • EP4390874A1 patent drawingFigure 1
  • EP4390874A1 patent drawingFigure 2
  • EP4390874A1 patent drawingFigure 3

AI summary

An information processing program causes a computer to execute processing including: acquiring a video; specifying a first region that includes an object included in the video, a second region that includes a person included in the video, and a relation that identifies an interaction between the object and the person, by analyzing the acquired video; determining whether or not the person included in the second region performs abnormal behavior, based on the specified first region and the specified relation; and notifying an alert related to appearance of the person who performs the abnormal behavior in a case of determining that the person performs the abnormal behavior.