Adaptive Background Model for Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional object detection methods fail to simultaneously detect stationary and moving objects while distinguishing them from backgrounds with changes, such as ambient light shifts or continuous changes like swaying trees and ripples of water, due to difficulties in distinguishing similar pixel variations.
Innovation Solution
An object detecting apparatus and method that determines pixel states based on temporal properties, adaptively generates a background model using the degree of similarity between pixel characteristics and a preceding background model, and judges objects based on this similarity, allowing for accurate detection of stationary or moving targets amidst changing backgrounds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If the background image is updated one after another by a simple addition method, then it is not necessary to prepare an accurate background image and slow background changes do not become an issue, but when the target object stands stationary or moves slowly, it becomes difficult to detect the target object because the target object is updated into the background image
Solution Approach 1:
The patent applies dynamics by making the background update mechanism adaptive rather than static. The system dynamically adjusts whether to update the background model based on detected object motion. When objects are detected as stationary or slowly moving, the background update is suppressed or reduced, preventing these objects from being incorrectly incorporated into the background model. This dynamic control resolves the contradiction between adapting to background changes and maintaining detection accuracy for stationary objects.
Solution Approach 2:
The patent changes the parameter of background update frequency and intensity based on object motion characteristics. By monitoring object velocity and positional changes over time, the system adjusts the background model update rate accordingly. For stationary or slow-moving objects, the update parameter is reduced or paused, while for fast-moving objects, normal background updates continue. This parameter adaptation resolves the technical contradiction.
2Measurement precision
If the variance value method is used to distinguish pixel states, then the target object can be detected even when standing stationary or moving slowly, but objects other than persons that stand stationary cannot be distinguished from the background
Solution Approach 1:
The patent applies universality by creating an object detection framework that works for all object types, not just persons. The system uses motion analysis and temporal consistency checks that are applicable to any stationary or moving object regardless of category. This universal approach distinguishes stationary objects from background while maintaining the ability to detect various object types, resolving the limitation of person-specific detection.
Solution Approach 2:
The patent segments the detection process into multiple stages: initial detection using variance analysis, followed by classification and verification stages. This segmentation allows the system to first identify potential stationary objects through variance methods, then apply additional criteria to distinguish them from background elements. The multi-stage approach enables versatile detection across different object types while maintaining accuracy.
3Measurement precision
If the background model is generated assuming the position and shape of detected object regions do not change, then person detection is improved, but objects whose position and size change such as swaying trees and ripples of water become an issue
Solution Approach 1:
The patent applies dynamics by making the background model adaptive to objects with changing position and size. Instead of assuming fixed position and shape, the system dynamically tracks and accommodates objects whose boundaries change over time, such as swaying trees or water ripples. This dynamic modeling approach maintains detection accuracy for persons while extending versatility to handle objects with variable geometry.
Solution Approach 2:
The patent changes the parameters of the background model to accommodate objects with varying position and size. The system adjusts model parameters such as region boundaries, shape descriptors, and temporal windows based on observed object behavior. This parameter adaptation allows the system to handle both stationary objects and objects with changing characteristics, resolving the technical contradiction.
Data Source
AI summary
An object detecting apparatus and method includes a pixel state determining unit that derives variance value for temporal properties of pixel characteristics of an input image, background model generating unit that adaptively generates a background model from characteristics in the characteristic storing unit and characteristic storing unit for background model generation using the characteristic distance and the pixel state determined as conditions, and an object judging unit that judges an object based on a characteristic distance indicative of a degree of similarity between a generated background model and pixel characteristics of an input image.