Image Processor Background Subtraction for Moving Fields of View
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing algorithms struggle to accurately discriminate between objects of interest and background regions, especially in scenarios where the field of view is not stationary, due to their computational intensity and requirement for multiple passes of individual frames.
Innovation Solution
An image processor that estimates and updates the background signal by shifting the field of view, using a high frame rate to overlap regions across frames, and applies a thresholding process to enhance foreground objects while suppressing background noise, utilizing a low-pass filter to refine the output.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If theoretical algorithmic solutions with multiple passes and feature detection are used, then discrimination accuracy between foreground objects and background is improved, but computational complexity and processing time increase significantly
Solution Approach 1:
The patent divides the image processing task into two distinct stages: background estimation using a simplified model, and foreground detection using the estimated background. This segmentation allows each stage to use appropriately complex methods, resolving the contradiction between overall accuracy and computational complexity.
Solution Approach 2:
The patent extracts and removes the background component from the image processing task by estimating it separately using a simplified model. This extraction allows the foreground detection to focus only on the remaining signal, improving accuracy without requiring complex processing of the entire image.
2Measurement precision
If multiple passes of individual frames with feature detection are used, then discrimination accuracy is improved, but processing speed and real-time performance deteriorate
Solution Approach 1:
The patent performs preliminary background estimation using a simplified model before conducting the actual foreground detection. This preliminary action removes the need for multiple passes through the data, as the background is already estimated and can be subtracted in a single operation, thereby improving processing speed.
Solution Approach 2:
The patent replaces complex mechanical-style multiple-pass processing with a more efficient computational approach: estimate the background once using a simple model, then subtract it in a single pass. This substitution eliminates the need for repeated mechanical-style iterations while maintaining discrimination accuracy.
3Measurement precision
If RANSAC and tracking algorithms are implemented, then consensus between salient points is improved, but computational intensity and resource requirements increase
Solution Approach 1:
The patent uses a simple, computationally inexpensive background estimation model that can be quickly computed and discarded after use. This disposable approach replaces expensive, energy-intensive algorithms like RANSAC and tracking, achieving sufficient consensus accuracy without the high computational energy cost.
4Productivity
If high frame rate processing is used to overlap regions across frames, then real-time performance is improved, but data storage requirements increase
Solution Approach 1:
The patent changes the parameter of interest from storing entire image frames to storing only the estimated background model parameters. This parameter transformation allows high frame rate processing with minimal storage requirements, as only the compact background model needs to be retained between frames.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method comprising: receiving a first image captured on an image sensor; obtaining a scene estimate for the first image, wherein the scene estimate comprises an estimate of the background signal for a region of space being imaged onto the sensor; storing the scene estimate in memory; receiving a plurality of successive images, wherein for each one of the successive images, the field of view is shifted with respect to the preceding image and overlaps partially with that of the preceding image; for each one of the successive images: retrieving the scene estimate from memory; subtracting the scene estimate from the image to obtain a background-subtracted image for output; updating the scene estimate based on the signal detected at the location on the image sensor to which the region of space is being imaged; and replacing the scene estimate stored in memory with the updated scene estimate.