Background Modeling for PTZ Video Streams After Camera Movement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional background modeling for cameras with movement capabilities, such as PTZ cameras, struggles to accurately distinguish between camera movements and scene changes, leading to incorrect categorization of static scene parts as foreground, and is limited by processing power constraints.
Innovation Solution
A hybrid background modeling approach that combines continuous updating of the background model during minimal movement with a quick reset using image segmentation and/or object detection algorithms upon camera stop, allowing for efficient power usage and accurate background identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Power
If conventional background modelling based on motion analysis is used for PTZ cameras, then the processing power requirement is reduced, but the accuracy of background identification deteriorates due to inability to distinguish camera movement from scene changes
Solution Approach 1:
The patent implements a dynamic background modelling approach that adapts to camera movement states. The system switches between conventional motion-based modelling during movement and AI-based segmentation during stationary periods, optimizing both power consumption and accuracy according to real-time camera state
Solution Approach 2:
The patent changes the modelling parameter from continuous motion analysis to stationary-frame AI segmentation based on camera state. By detecting when the camera is stationary and switching to a different modelling approach (AI-based object detection and segmentation), the system achieves high accuracy without continuous high-power processing
2Measurement precision
If AI-based image segmentation and object detection algorithms are continuously applied, then the background identification accuracy is improved, but the power consumption increases significantly
Solution Approach 1:
The patent applies AI-based segmentation periodically only when the camera is stationary, rather than continuously. The system monitors camera movement and triggers the computationally intensive AI algorithms only during stationary periods, reducing overall power consumption while maintaining accuracy when needed
Solution Approach 2:
The patent dynamically adjusts the processing approach based on camera state. During movement, the system uses low-power conventional methods; during stationary periods, it switches to high-accuracy AI-based methods. This dynamic adaptation optimizes the balance between power consumption and identification accuracy
3Speed
If the background model is continuously updated during camera movement, then the responsiveness is improved, but the reliability of background categorization deteriorates due to incorrect foreground identification
Solution Approach 1:
The patent implements a dynamic update strategy that pauses conventional background modelling during camera movement and resumes only when the camera is stationary. This prevents unreliable updates during movement while maintaining responsiveness through quick resumption of updates once stability is detected
Solution Approach 2:
The patent performs preliminary camera movement detection before resuming background model updates. By detecting stationary state in advance and then triggering the reset and update process, the system ensures that updates only occur when reliable, preventing incorrect categorization while maintaining overall responsiveness
Data Source
AI summary
The present disclosure relates to a method of background modelling for a video stream acquired by a camera having movement capabilities. The method comprises: acquiring a video stream; repeatedly updating a background model for the video stream by analyzing changes in the sequence of image frames and categorizing image areas in the image frames which do not change over time as background; detecting camera movement; performing a reset of the background model by: applying an image segmentation and/or object detection algorithm to identify at least foreground objects; after having performed the reset of the background model, returning to repeatedly updating the background model for the video stream by analyzing changes in the sequence of image frames and categorizing image areas in the image frames which do not change over time as background. The disclosure further relates to an image processing system.