Aggregated Background Subtraction for Action Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing action classification systems in video surveillance are computationally intensive and require powerful GPUs, making them costly and prone to latency, especially when run on remote servers or in the cloud, which limits their accessibility and real-time capabilities.
Innovation Solution
A light-weight action classification system that uses aggregated background subtraction (ABS) images, trained with a neural network, to classify actions by objects, reducing the need for continuous GPU access and improving performance under varying conditions like illumination changes and motion.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning driven object detection solutions are used to classify actions by objects, then classification accuracy is improved, but computational complexity and cost increase significantly
Solution Approach 1:
The patent segments the video processing task into two distinct components: (1) generating aggregated background subtraction images that capture motion patterns, and (2) classifying actions based on these simplified representations. This segmentation allows the complex deep learning classifier to operate on reduced data with fewer computational requirements while maintaining accuracy.
Solution Approach 2:
The patent introduces aggregated background subtraction images as an intermediary representation between raw video frames and the action classifier. This intermediary format extracts essential motion information while eliminating redundant background data, enabling the classifier to achieve high accuracy with reduced computational complexity.
2Power
If GPU-based deep learning algorithms are deployed on remote servers or cloud platforms, then processing power is improved, but latency in event notifications increases
Solution Approach 1:
The patent performs preliminary action by pre-processing video frames into aggregated background subtraction images that highlight motion patterns. This pre-processing step creates a simplified representation that requires less computational power for classification, enabling faster processing and reduced latency when deploying on remote servers or cloud platforms.
3Productivity
If traditional background subtraction methods are used to process video frames, then processing speed is improved, but sensitivity to illumination changes and environmental conditions worsens
Solution Approach 1:
The patent merges multiple frame differences into aggregated background subtraction images by accumulating absolute differences across multiple frames. This merging process reinforces consistent motion patterns while suppressing random noise from environmental factors like illumination changes, thereby improving reliability without sacrificing processing speed.
Solution Approach 2:
The patent maintains continuity of useful action by continuously accumulating frame differences over multiple frames to build the aggregated background subtraction image. This continuous accumulation process preserves motion information while filtering out transient environmental disturbances, improving reliability while maintaining processing efficiency.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for classifying actions of an object. The methods, systems, and apparatus include actions of: obtaining frames of video including an object of interest; determining a type of action of the object in each of the frames of video; determining a group of frames from the frames of video based on the type of action; determining an aggregated background subtraction (ABS) image based on adjacent frames of the group of frames; generating a training set that includes labeled ABS images including the ABS image; and training an action classifier using the training set.