Video Rotation Estimation Using Homography and Loss Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video rotation estimation methods suffer from low precision due to the lack of 2D spatial point information, or high computational complexity and instability in diverse motion scenarios, particularly when translation is minimal.
Innovation Solution
A method involving calculating a global rotation initial value and intrinsic matrix using matching relationships between frames, followed by homography matrix calculation and optimization to minimize a loss function, improving precision and reducing computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If rotation average algorithm is used to solve global rotation based on relative rotation between multiple two-frame images, then computation speed is fast, but estimation precision is not high because 2D spatial points information is not used
Solution Approach 1:
The patent merges two different algorithmic approaches: the rotation average algorithm (which provides fast computation) and the homography-based algorithm using 2D spatial points (which provides high precision). By combining these methods in a unified optimization framework that processes multiple frames simultaneously, the system achieves both computational efficiency and estimation accuracy.
Solution Approach 2:
The patent transitions from analyzing only relative rotation between two-frame images to utilizing 2D spatial points across multiple frames. This dimensional expansion allows the system to leverage spatial information from multiple dimensions (temporal and spatial) to improve estimation precision while maintaining computational feasibility through optimized processing.
2Measurement precision
If SFM algorithm or SLAM algorithm is used to directly solve poses and coordinates of 3D spatial points, then estimation precision is high, but computation speed is slow and robustness cannot be guaranteed
Solution Approach 1:
The patent extracts and utilizes only the essential components needed for rotation estimation from the complex SFM/SLAM frameworks. By focusing specifically on homography matrix calculation and optimization for rotation purposes, the system achieves high precision without the excessive computational burden of complete pose and 3D point reconstruction.
Solution Approach 2:
Instead of performing complete SFM or SLAM processing, the patent applies a partial action approach by computing homography matrices and optimizing rotation parameters directly from 2D spatial points. This selective processing achieves sufficient precision for rotation estimation while significantly reducing computational complexity and improving speed.
3Measurement precision
If SFM algorithm or SLAM algorithm is used to solve poses, then estimation precision is high, but the algorithm is prone to failure in many motion modes and scenarios
Solution Approach 1:
The patent changes the fundamental parameters and assumptions of the estimation algorithm. Instead of relying on SFM/SLAM assumptions of significant translation and 3D structure, the system uses homography-based optimization that works effectively with 2D spatial points and can handle diverse motion modes including pure rotation, picture stillness, and scenarios with minimal translation.
Solution Approach 2:
The patent creates a universal rotation estimation method that functions reliably across diverse motion scenarios. The homography-based approach with optimization can handle various motion modes (pure rotation, translation, stillness, distant view) that cause SFM/SLAM algorithms to fail, providing robust and reliable estimation in all cases.
Data Source
AI summary
A method and apparatus for estimating the rotation of a video, and an electronic device and a storage medium. The method for estimating video rotation includes: calculating a global rotation initial value of a current frame of image and an intrinsic matrix of an image acquiring apparatus according to a first matching relationship; calculating a homography matrix between the current frame of image and each of M frames of images before and after the current frame of image according to a second matching relationship; setting a loss function according to the global rotation initial value and the intrinsic matrix, and minimizing the loss function by using an optimization method to obtain a final loss function; and calculating a global rotation optimization value of the current frame of image according to the final loss function and multiple homography matrices.