Staring Sensor Pixel Variance Estimation for Jitter Mitigation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Camera jitter in staring sensors leads to false detections and reduced detection sensitivity due to scene-induced clutter, which conventional background suppression techniques struggle to mitigate, especially in outdoor applications with dynamic backgrounds.
Innovation Solution
A method that involves capturing a reference image frame and a current image frame, generating a raw difference frame, scaling it with a spatial or hybrid standard deviation frame to create a normalized difference frame, and applying a detection threshold filter to accurately detect changes while reducing jitter-induced artifacts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional background suppression techniques are used to detect changes in staring sensor scenes, then detection sensitivity is reduced and false detections increase due to scene-induced clutter from camera jitter, but implementing sub-pixel registration to mitigate jitter is not computationally feasible at sensor frame rates and not robust to pixel defects
Solution Approach 1:
The patent segments the image processing task by computing pixel variances independently for each pixel location across multiple frames. This divides the complex registration problem into simpler per-pixel statistical computations that can be performed at full frame rates without requiring computationally intensive sub-pixel registration algorithms.
Solution Approach 2:
The patent performs preliminary statistical analysis by computing pixel variances from a sequence of frames before change detection. This pre-computation of variance maps provides a foundation for later change detection that is robust to jitter, avoiding the need for complex real-time registration during the actual change detection process.
2Measurement precision
If background suppression is applied to detect transient events, then changes can be detected, but pixels in high scene gradient regions change substantially with increased jitter leading to clutter false detections
Solution Approach 1:
The patent applies local quality by computing pixel variances independently for each pixel location, allowing the detection threshold to adapt to local scene characteristics. Pixels in high gradient regions naturally exhibit higher variances due to jitter, and this is accounted for by using the locally-computed variance as a basis for change detection, making the detection precision location-dependent and robust to jitter-induced clutter.
Solution Approach 2:
The patent changes the parameter used for change detection from raw intensity differences to variance-normalized differences. By dividing the intensity difference by the square root of the computed variance, the detection becomes invariant to the magnitude of jitter-induced variations, thereby eliminating clutter false detections in high gradient regions while maintaining sensitivity to true changes.
3Adaptability or versatility
If filters are used to predict intensity values in the presence of dynamic backgrounds, then repetitive patterns can be learned, but these adaptive techniques still fail in the presence of camera jitter
Solution Approach 1:
The patent introduces pixel variance as an intermediary parameter that mediates between the raw intensity values and the change detection decision. This variance intermediary captures the effect of jitter statistically, allowing the system to adapt to dynamic backgrounds while remaining robust to camera jitter, as the variance normalization removes the jitter component from the detection criterion.
Data Source
AI summary
A technique for detecting changes in a scene perceived by a staring sensor is disclosed. The technique includes acquiring a reference image frame and a current image frame of a scene with the staring sensor. A raw difference frame is generated based upon differences between the reference image frame and the current image frame. Pixel error estimates are generated for each pixel in the raw difference frame based at least in part upon spatial error estimates related to spatial intensity gradients in the scene. The pixel error estimates are used to mitigate effects of camera jitter in the scene between the current image frame and the reference image frame.