Video Human Detection via Segmented Map and Selective Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing video surveillance systems face challenges in accurately detecting and counting humans in crowded scenes due to high computational costs, viewpoint dependency, and the need for extensive training, which limits their effectiveness in real-world applications such as public places and busy areas.

Innovation Solution

A method that uses frame differencing to detect moving pixels, builds a dynamic background model, and tracks foreground blobs to determine human presence, employing a calibrated human shape model to estimate human probability across the image, allowing for crowd density analysis and event detection without requiring extensive training.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If detection-based approaches are used to identify each individual human target, then detection accuracy is improved in lightly crowded scenarios, but computational cost increases significantly

Engineering Contradiction:
Improvedetection accuracyVSAvoidcomputational cost
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the detection process into two distinct stages: (1) a fast map-based approach that processes the entire image to generate initial candidate locations with low computational cost, and (2) a selective detection-based approach that applies computationally intensive individual target detection only to promising candidate regions identified in stage 1. This segmentation resolves the contradiction by distributing computational load across different processing stages.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial detection-based action by performing detailed individual target detection not on the entire image, but only on selected candidate regions identified by the map-based approach. This partial application of the computationally expensive detection method maintains accuracy where needed while significantly reducing overall computational cost compared to applying detection-based methods to the complete image.

Inventive Principle:
Principle #16Partial or excessive action

2Use of energy by moving object

If map-based approaches are used to estimate human count from image features, then computational cost is reduced, but detection accuracy deteriorates in crowded scenes and actual locations of individuals become unavailable

Engineering Contradiction:
Improvecomputational costVSAvoiddetection accuracy
Core Design Contradiction:
Use of energy by moving objectVSMeasurement precision

Solution Approach 1:

The patent segments the detection process into two distinct stages: (1) a fast map-based approach that processes the entire image to generate initial candidate locations with low computational cost, and (2) a selective detection-based approach that applies computationally intensive individual target detection only to promising candidate regions identified in stage 1. This segmentation resolves the contradiction by distributing computational load across different processing stages.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary component - the map-based approach - that acts as a bridge between the full image and the detection-based method. This intermediary generates candidate location maps that guide subsequent detailed detection, enabling the system to benefit from both the efficiency of map-based methods and the accuracy of detection-based methods without fully committing to either approach alone.

Inventive Principle:
Principle #24Intermediary (Mediator)

3Loss of information

If detection-based approaches are used to identify each individual human target, then individual locations become available for crowd density computation, but the approach becomes viewpoint dependent and requires large human image size

Engineering Contradiction:
Improveindividual location informationVSAvoidviewpoint dependency
Core Design Contradiction:
Loss of informationVSAdaptability or versatility

Solution Approach 1:

The patent segments the detection process into two distinct stages: (1) a fast map-based approach that processes the entire image to generate initial candidate locations with low computational cost, and (2) a selective detection-based approach that applies computationally intensive individual target detection only to promising candidate regions identified in stage 1. This segmentation resolves the contradiction by distributing computational load across different processing stages.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent enhances viewpoint adaptability by integrating two approaches that complement each other's strengths. The map-based approach provides viewpoint-independent candidate generation, while the detection-based approach refines individual locations when needed. This multi-functional system can adapt to different viewpoints and crowd densities by selectively applying appropriate methods.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP2895986B1Methods, devices and systems for detecting objects in a video
Publication Date: 2019.03.06 AVIGILON FORTRESS
  • EP2895986B1 patent drawingFigure 1
  • EP2895986B1 patent drawingFigure 2
  • EP2895986B1 patent drawingFigure 3A

AI summary

Methods, devices and systems for performing video content analysis to detect humans or other objects of interest a video image is disclosed. The detection of humans may be used to count a number of humans, to determine a location of each human and/or perform crowd analyses of monitored areas.