Stationary Object Detection via Multi-mode Background Modelling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video processing techniques face challenges in robustly detecting stationary foreground objects and distinguishing them from slowly moving or intermittently moving objects, often losing track due to occlusion or noise, and struggle to effectively handle both cases simultaneously.
Innovation Solution
The Multi-mode Background Modelling (MMBM) approach uses 'age' and 'activity count' of video scene elements, in conjunction with mode models and thresholds, to determine stationary objects by calculating representative age and activity measures and comparing them to threshold values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the system is made more sensitive to changes in slowly moving objects, then detection of slowly moving objects improves, but the system loses track of stationary objects due to occlusion or noise
Solution Approach 1:
The system dynamically adjusts its detection sensitivity based on the temporal characteristics of detected objects. By maintaining multiple mode models with different activity count thresholds, the system can adaptively switch between high sensitivity (for detecting slowly moving objects) and low sensitivity (for maintaining stable tracking of stationary objects), resolving the contradiction between detection precision and tracking reliability
Solution Approach 2:
The invention changes the parameter of detection sensitivity by using activity count thresholds derived from mode models. Objects are classified as stationary or moving based on their activity count relative to the mode model thresholds, allowing the system to adjust its response to different object types without fixed sensitivity settings, thereby balancing detection precision and tracking stability
2Ease of operation
If the system uses a single technique to detect both stationary objects and slowly moving objects, then operational simplicity improves, but the system cannot effectively handle both cases simultaneously
Solution Approach 1:
The system achieves universality by using a single multi-mode background modelling framework that can handle both stationary and slowly moving objects. The mode models maintain multiple representations of the background with different activity counts, allowing the same detection mechanism to adaptively respond to different object types based on their temporal characteristics, thus providing both operational simplicity and versatility
3Measurement precision
If the system uses background subtraction with reference frames or mode models, then foreground/background segmentation improves, but the system struggles to distinguish truly stationary objects from objects moving slowly or intermittently
Solution Approach 1:
The system performs preliminary classification by comparing activity counts against mode model thresholds before final object classification. This preliminary action using temporal characteristics helps pre-sort objects into stationary or moving categories, reducing the complexity of subsequent classification while maintaining high segmentation accuracy
Data Source
AI summary
Disclosed is a method 1201 of processing a video stream, the method comprising the steps of determining 1230 a representative age measure from a model for a visual element from the video stream, determining 1250 a representative activity count measure from the model, establishing a functional relationship between the representative activity count measure and the representative age measure, comparing 1240 the functional relationship to a threshold value, and determining 1260 if the visual element is stationary, based on the result of the comparing step.