Compressive sensing block sizing for video anomaly detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video surveillance systems require significant network and computational resources for real-time monitoring and human operators to detect anomalies, leading to high costs and potential missed detections due to operator fatigue and errors.
Innovation Solution
The method employs compressive sensing to divide video data into spatio-temporal blocks of varying sizes based on distance information, allowing for anomaly detection without reconstructing the video data, thereby reducing computational and network requirements and eliminating the need for human operators.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional compression standards (H.264) are used for real-time video transmission, then video quality is maintained, but network bandwidth and computational resources are excessively consumed
Solution Approach 1:
The patent extracts only the essential features and measurements from video data using compressive sensing, rather than transmitting and processing complete video frames. This allows anomaly detection to be performed on compressed measurements, significantly reducing network bandwidth consumption while maintaining detection reliability
Solution Approach 2:
The system performs partial action by detecting anomalies without fully reconstructing video frames from compressed measurements. This partial reconstruction approach consumes far fewer computational resources compared to conventional full decompression while still enabling effective anomaly detection
2Measurement precision
If human operators review video data continuously to detect anomalies, then detection accuracy is maintained, but operational costs increase and operator fatigue causes missed detections
Solution Approach 1:
The system implements self-service by using automated anomaly detection algorithms that process compressed video measurements without requiring continuous human operator intervention. The system autonomously identifies anomalies, eliminating operational costs and fatigue-related errors while maintaining consistent detection accuracy
Solution Approach 2:
The patent replaces the mechanical system of human visual inspection with an automated computational system that processes compressive measurements. This substitution eliminates the need for human operators while providing consistent, fatigue-free anomaly detection capability
3Device complexity
If video data is divided into fixed-size blocks for compressive sensing, then processing is simplified, but detection accuracy decreases for objects at varying distances
Solution Approach 1:
The patent implements dynamic block sizing where the size of video blocks is adjusted based on the distance to objects in the scene. Objects farther away are represented by smaller blocks while closer objects use larger blocks, optimizing the representation for objects at varying distances and improving detection accuracy without excessive complexity
Solution Approach 2:
The system applies local quality by using different block sizes in different regions of the video frame based on distance information. This allows the processing complexity to be adapted locally to match the actual content and distance requirements of each region, improving overall detection precision
Data Source
AI summary
A method and a system for using compression sensing to provide low data rate transmission and low computational complexity to determine anomalies in video data obtained by a video camera or other motion detection device. The video data is divided into varying sized video blocks based on an anticipated size of objects of interest within the video, and based on a distance between a video camera and the objects of interest. Features are extracted from the video data of each block to detect anomalies if a feature vector is outside of an “allowed range.” By utilizing varying sized video blocks, anomalies are more effectively and efficiently detected in the video data.