Robust Video Change Detection Using Orthogonal Transform
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video surveillance and analytics systems face challenges in robustly detecting changes in video streams due to susceptibility to brightness and contrast variations, noise, and environmental disturbances, particularly in thermal camera images, which can lead to false alarms and reduced detection accuracy.
Innovation Solution
The system employs orthogonal transformations, such as Discrete Cosine Transform (DCT) or Fast Wavelet Transform (FWT), on non-overlapping blocks of video and reference images, zeroing DC components and applying local scaling factors, followed by determining the infinity-norm difference to compare with thresholds, thereby focusing on local patterns and textures while being insensitive to random noise.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional change detection methods are used, then the system can identify changes in video streams, but it produces false alarms due to susceptibility to brightness and contrast variations and noise
Solution Approach 1:
The patent divides the video image and reference image into non-overlapping blocks, with each block in the video image corresponding to a block in the reference image. This segmentation allows the system to process and compare local regions independently, reducing the impact of global brightness and contrast variations while maintaining sensitivity to local changes.
Solution Approach 2:
The patent applies an orthogonal transformation (such as DCT or FWT) to each block, converting the image data from the spatial domain to a transformed domain. This parameter transformation enables the system to zero out DC components and apply local scaling factors, making the change detection robust to brightness and contrast variations while preserving important local patterns and textures.
2Measurement precision
If the system is made more sensitive to detect all changes, then detection capability improves, but false alarms increase due to noise and environmental disturbances
Solution Approach 1:
The patent extracts and removes the DC component (average value) from each orthogonally transformed block by setting it to zero. This extraction eliminates the impact of brightness and contrast variations, allowing the system to focus on detecting only the AC components that represent actual changes in the scene, thereby reducing false alarms from noise and environmental disturbances.
Solution Approach 2:
The patent applies different processing to different blocks based on their local characteristics. By applying orthogonal transformation and DC component removal to each block individually, the system adapts to local variations in the image, maintaining high sensitivity to real changes while being insensitive to noise and environmental factors in each local region.
3Device complexity
If the system processes images in the original pixel domain, then computation is simple, but it cannot distinguish between interesting changes and non-interesting background variations
Solution Approach 1:
The patent transforms the image data from the spatial domain to a transformed domain using orthogonal transformation (DCT or FWT). This dimensional change allows the system to represent image information in terms of frequency components, where DC components represent average intensity and AC components represent variations. This transformation enables the system to distinguish between interesting changes and non-interesting background variations more effectively.
Data Source
AI summary
A system and method for detecting change in a video stream may include receiving, by a processor, a video image from a video stream and a reference image. The processor may divide the video image and the reference image into non-overlapping blocks, and each block in the video image may correspond to a block in the reference image. The processor may perform an orthogonal transformation on a corresponding pair of blocks in the video image and the reference image. DC (direct current, or, average value) components in the orthogonally transformed pair of blocks in the video image and the reference image may be zeroed. An infinity-norm difference between coefficient values of the corresponding pair of transformed and modified blocks in the video image and the reference image may be determined and compared to a threshold.