Surveillance Object Classification via Dynamic Background Suppression

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing surveillance security systems often mistakenly classify uninteresting dynamic background objects as foreground objects, leading to unnecessary alerts and increased computational costs due to pixel-level foreground segmentation algorithms.

Innovation Solution

The system classifies all objects as dynamic background initially and reclassifies them as foreground based on motion and appearance cues at the object level, using unique identifiers and tracking positions over time to suppress alerts for uninteresting objects.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If pixel-level foreground segmentation algorithms are used to identify objects, then foreground objects can be detected, but uninteresting dynamic background objects are mistakenly classified as foreground objects

Engineering Contradiction:

Improveobject classification accuracyVSAvoidfalse positive alerts

Core Design Contradiction:

Measurement precisionVSObject-generated harmful factors

Solution Approach 1:

The patent divides the classification process into two levels: pixel-level segmentation to identify potential foreground regions, and object-level analysis to classify those regions as either foreground objects or dynamic background. This hierarchical segmentation resolves the contradiction by separating detection from classification, allowing pixel-level algorithms to maintain sensitivity while object-level analysis eliminates false positives.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Instead of trying to distinguish foreground from background at the pixel level, the patent inverts the approach by initially classifying all detected objects as dynamic background, then promoting only those that exhibit foreground-like behavior. This inversion resolves the contradiction by assuming innocence until proven guilty, reducing false positives while maintaining detection accuracy.

Inventive Principle:

Principle #13The other way round (Inversion)

2Object-generated harmful factors

If pixel-level foreground segmentation algorithms are modified to exclude dynamic background objects, then false positive alerts are reduced, but some foreground objects are mistakenly classified as background objects

Engineering Contradiction:

Improvefalse positive alertsVSAvoidforeground object detection accuracy

Core Design Contradiction:

Object-generated harmful factorsVSMeasurement precision

Solution Approach 1:

The patent segments the classification task into pixel-level detection and object-level verification. The pixel-level algorithm maintains its sensitivity to all moving objects, while the subsequent object-level analysis stage applies criteria to distinguish foreground from dynamic background, preventing both false positives and false negatives.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary classification of all objects as dynamic background, then applies promotion criteria to identify true foreground objects. This preliminary action ensures that no foreground objects are prematurely excluded, while the promotion stage systematically identifies true positives based on multiple indicators.

Inventive Principle:

Principle #10Preliminary action

3Measurement precision

If pixel-level foreground segmentation algorithms are modified to improve classification accuracy, then object identification improves, but computational cost increases

Engineering Contradiction:

Improveobject classification accuracyVSAvoidcomputational energy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent segments the computational workload into two stages: a fast pixel-level segmentation that identifies potential objects, and a more computationally intensive object-level analysis that is applied only to detected objects rather than all pixels. This segmentation reduces overall computational energy while maintaining classification accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies the more computationally intensive classification criteria partially - only to objects that are detected at the pixel level rather than to all pixels in the image. This partial application of the algorithm maintains accuracy for relevant objects while significantly reducing overall computational energy consumption.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS10861163B2System and method for identification and suppression of time varying background objects

Publication Date: 2020.12.08 TYCO FIRE & SECURITY GMBH

AI summary

A system and method for identification and suppression of time varying background objects is disclosed. A proposed surveillance security system includes an image analytics system and at least one surveillance camera. The image analytics system identifies objects within image data of a scene captured by the surveillance camera, and then analyzes the objects to determine whether each object is a foreground object or a dynamic background object. In examples, the image analytics system determines whether an object is a foreground object or a dynamic background object based upon movement and/or an appearance of the object. The surveillance security system does not send alerts to users of the system for objects determined to be dynamic background objects. When users request objects of interest, the dynamic background objects are also excluded from lists of objects sent in reply messages to the users.