Active Attentional Sampling for Video Background Subtraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current background subtraction algorithms face significant computation time challenges, especially with increasing video sizes, making real-time applications impractical due to heavy computational loads, despite efforts to optimize algorithms and utilize parallel processing or selective sampling.
Innovation Solution
An active attentional sampling method that utilizes temporal, spatial, and frequency properties of input videos to generate a foreground probability map and sampling mask, allowing for selective pixel-wise background subtraction, thereby reducing computational complexity and accelerating the background subtraction process.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-based probabilistic model methods are used for background subtraction, then detection performance is improved, but computation time increases significantly
Solution Approach 1:
The patent divides the image into multiple regions of interest (ROIs) based on foreground probability maps, and applies different processing strategies to different regions. High-probability regions undergo full pixel-wise background subtraction, while low-probability regions use coarser sampling or are skipped entirely, thereby reducing overall computation time while maintaining detection performance in critical areas.
Solution Approach 2:
The patent applies different levels of processing quality to different spatial locations based on their foreground probability. Regions with high foreground probability receive more computationally intensive processing (full pixel-wise analysis), while regions with low probability receive lighter processing (coarse sampling or skipping), optimizing the trade-off between detection accuracy and computation time.
2Productivity
If selective sampling methods are used to reduce computation time, then processing speed is improved, but small objects may be missed
Solution Approach 1:
The patent dynamically adjusts the sampling strategy for each frame based on the foreground probability map and detected objects from previous frames. The sampling rate and region selection change adaptively according to the current scene content, allowing the system to maintain high detection accuracy for small objects when they are present while achieving speedup during static scenes.
Solution Approach 2:
The patent uses feedback from foreground detection results and probability maps to guide subsequent sampling decisions. Detected objects and their probabilities inform which regions require intensive sampling versus which regions can use coarser sampling, creating a closed-loop system that maintains reliability while improving speed.
3Productivity
If grid pattern sampling is used for background subtraction, then computation is reduced, but redundant operations remain
Solution Approach 1:
The patent applies partial action by performing background subtraction only on necessary regions identified by foreground probability maps, rather than uniformly processing the entire image. This selective approach eliminates redundant operations in background-dominated regions while maintaining necessary processing in regions with potential foreground content.
Data Source
AI summary
An active attentional sampling technology for accelerating background subtraction from input videos, more specifically, an active attentional sampling technology for accelerating background subtraction by removing background region from the input video and then applying foreground probability map and sampling mask according to temporal property, spatial property and frequency property of the input video in favor of the pixel-wise background subtraction algorithm is provided. The background subtraction is accomplished by generating active attentional sampling mask for input video and then processing each frame of the input video only for regions corresponding to the sampling mask, which renders the background subtraction be much accelerated. That is, the present invention successfully speeds up pixel-wise background subtraction methods approximately 6.6 times without deteriorating detection performance. Therefore, according to the present invention, real-time detection with full-HD video is successfully achieved through various conventional background subtraction algorithms.