Risky Behavior Detection via Silhouette Distance Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional surveillance systems for detecting risky behavior require operator analysis and often infringe on individuals' privacy by capturing video images, leading to unsatisfactory surveillance quality and comfort issues.

Innovation Solution

A detection network utilizing an artificial neural network to analyze temporal sequences of silhouette distances from topographic views, eliminating the need for video image capture and enabling autonomous detection of risky behavior without operator intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If video images are captured to detect risky behavior, then detection accuracy is improved, but privacy protection deteriorates

Engineering Contradiction:
Improvedetection accuracyVSAvoidprivacy infringement
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent extracts only the essential information needed for detection (silhouette and distance data) from the complete video image, discarding all non-essential visual information that would compromise privacy. This allows detection to proceed with minimal data collection, resolving the contradiction between accuracy and privacy.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

Instead of using actual video images, the system creates simplified representations (silhouettes) that capture only the necessary spatial and temporal information for behavior analysis. This copy contains sufficient detection data while eliminating privacy-sensitive details.

Inventive Principle:
Principle #26Copying

2Adaptability or versatility

If operator analysis is used to detect risky behavior, then detection flexibility is improved, but productivity deteriorates

Engineering Contradiction:
Improvedetection flexibilityVSAvoidsurveillance efficiency
Core Design Contradiction:
Adaptability or versatilityVSProductivity

Solution Approach 1:

The system enables automatic detection through algorithmic analysis of silhouette sequences, allowing the surveillance system to serve itself without human intervention. The artificial intelligence processes the data autonomously, maintaining flexibility while dramatically improving productivity and efficiency.

Inventive Principle:
Principle #25Self-service

3Area of stationary object

If video cameras are deployed to monitor living areas, then surveillance coverage is improved, but ease of operation deteriorates

Engineering Contradiction:
Improvesurveillance coverageVSAvoiduser comfort
Core Design Contradiction:
Area of stationary objectVSEase of operation

Solution Approach 1:

The system extracts only silhouette and distance information from the monitored space, removing the need for comprehensive video imaging. This minimal data collection approach maintains surveillance coverage while significantly improving user comfort and acceptance by eliminating intrusive visual monitoring.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3848916B1Device for detecting hazardous behaviour of at least one person, associated detection method and detection network
Publication Date: 2023.11.01 KXKJM
  • EP3848916B1 patent drawingFigure 1
  • EP3848916B1 patent drawingFigure 2

AI summary

The invention relates to a detection device comprising: - a depth sensor (28) configured to generate a plurality of topographic views of the detection field (24), each topographic view comprising topographic distances between the depth sensor (28) and points within the detection field (24), - a determination module (30) configured to determine a silhouette of the person (22) from at least one topographic view, the silhouette being defined by a subset of the topographic distances, referred to as silhouette distances, - a processing module (32) configured to determine the risky behavior of the person (22) by propagating at least one temporal sequence of silhouette distances via an artificial neural network. The artificial neural network is trained on a temporal sequence of silhouette distances corresponding to reference behaviors of the person (22).