Moving Object Detection via Homography Warping and Optical Flow
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting moving objects, especially small ones, are limited by their inability to handle videos from moving cameras and are prone to false positives due to image artifacts and varying backgrounds.
Innovation Solution
A method that involves extracting corner points, feature vectors, and corresponding points in consecutive images, applying homography and warping, calculating optical flow, and using low-pass filtering and thresholding to identify moving objects, even in videos captured by a moving camera.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If standard optical flow algorithms are used with a fixed background assumption, then detection simplicity is maintained, but detection reliability deteriorates when using moving cameras
Solution Approach 1:
The patent applies dynamics by transitioning from a static background assumption to a dynamic background model that adapts to camera motion. The system estimates camera movement between frames and uses this information to adjust the background model dynamically, allowing reliable detection even when the camera is moving. This resolves the contradiction by maintaining operational simplicity while improving reliability through adaptive background modeling.
2Measurement precision
If camera moves closer to small moving objects, then detection precision is improved, but image artifacts from compression increase
Solution Approach 1:
The patent converts the harmful effect of image compression artifacts into a beneficial signal by exploiting the fact that real moving objects also cause intensity changes between frames. The system uses temporal filtering and motion analysis to distinguish between artifacts caused by compression and genuine object motion, allowing precise detection of small objects even when the camera is close. This transforms the harmful compression artifacts into a detectable pattern that can be differentiated from actual object motion.
3Reliability
If low pass filtering and thresholding are applied to remove high frequencies, then false positives are reduced, but detection sensitivity may decrease
Solution Approach 1:
The patent applies local quality by using adaptive thresholding that adjusts sensitivity based on local image characteristics and motion patterns. Rather than applying a uniform threshold across the entire image, the system analyzes local regions and applies appropriate filtering strength to each area. This allows the system to reduce false positives in stable regions while maintaining high sensitivity in regions where actual object motion is detected, thus resolving the contradiction between reliability and sensitivity.
Data Source
AI summary
A method of detecting moving objects, including the following steps: extracting first and second images from a video of an observed space; detecting corner points in the images; using the points, extracting a feature vector; identifying corresponding points in the images; determining a homography to perform image registration between the images; applying a warping of the second image based on the homography to obtain a warped second image; calculating a difference in individual pixels between the warped second image and the first image to identify pixels that differ; obtaining a third image by comparing the second warped image against the first image; identifying a bounding box in the third image for each group of adjacent pixels; calculate the optical flow based on the first image and the warped second image; extracting from the optical flow a matrix corresponding to the magnitude of the optical flow.