Vehicle Video Stabilization Using Grid Motion Vectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing vehicle-mounted video stabilization methods suffer from low accuracy in motion estimation, leading to inadequate image stabilization processing effects.

Innovation Solution

A vehicle-mounted video stabilization method using a pre-trained neural network to estimate motion vectors in a grid manner, followed by smoothing and optical flow calculations to improve accuracy, and employing neural networks tailored to different vehicle speeds and lighting conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single motion matrix estimation method is used for the whole image, then the processing is simple, but the accuracy of motion estimation is low

Engineering Contradiction:
Improveaccuracy of motion estimationVSAvoidcomplexity of motion estimation method
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The image is divided into multiple grids, and each grid is processed independently to obtain local motion vectors. This segmentation approach allows the system to capture local motion variations across different regions of the image, significantly improving motion estimation accuracy compared to single global motion matrix methods.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from estimating a single global motion matrix to estimating multiple local motion vectors across a grid dimension. This dimensional expansion from one global parameter to multiple local parameters enables the system to represent complex motion patterns more accurately.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

2Measurement precision

If neural network is used for motion estimation, then the processing effect can be improved, but the computation amount increases

Engineering Contradiction:
Improveaccuracy of motion estimationVSAvoidcomputation amount
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The neural network processes each grid independently rather than the entire image at once. This segmentation reduces the computational burden on the neural network while maintaining accurate motion estimation through local feature analysis.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies neural network processing selectively to extract key motion features from each grid, rather than processing all pixel data comprehensively. This partial action approach achieves sufficient motion estimation accuracy with reduced computation.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4468256B1Vehicle-mounted video image stabilization method and apparatus, vehicle and storage medium
Publication Date: 2026.03.04 GREAT WALL MOTOR CO LTD
  • EP4468256B1 patent drawingFigure 1~2
  • EP4468256B1 patent drawingFigure 3~5
  • EP4468256B1 patent drawing

AI summary

The present application is applicable to the technical field of image processing, and provides a vehicle-mounted video stabilization method and apparatus, a vehicle, and a storage medium. The method includes: obtaining a vehicle-mounted video of a vehicle; inputting the vehicle-mounted video into a first neural network to obtain a plurality of grid-shaped first motion vector graphs, where the first motion vector graph is used to represent a corresponding offset vector of each grid of a next frame of image of every two adjacent frames of images with respect to each grid of a previous frame of image in the vehicle-mounted video; performing a smoothing processing on the first motion vector graph to obtain a processed second motion vector graph; obtaining a first motion optical flow graph by calculation according to the second motion vector graph, and performing a motion compensation on a frame image in the vehicle-mounted video according to the first motion optical flow diagram to obtain a video having stabilized images. This method improves an accuracy of motion estimation of the neural network on the video image, thereby improving a processing effect of the video having stabilized images.