Driving video generation method based on spatio-temporal factorization architecture and hybrid modulation

By employing a spatiotemporal factorization architecture and a hybrid modulation method for generating driving videos, the problem of insufficient multi-view information fusion was solved, resulting in more accurate autonomous driving scene generation and video prediction, and improving the stability and consistency of the generated results.

CN122093518BActive Publication Date: 2026-07-07JILIN UNIVERSITY
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN ยท China
Patent Type
Patents(China)
Current Assignee / Owner
JILIN UNIVERSITY
Filing Date
2026-04-27
Publication Date
2026-07-07

AI Technical Summary

Technical Problem

Existing methods for autonomous driving scene generation and video prediction suffer from insufficient correlation modeling in terms of multi-view, multi-time series and multi-modal information fusion, making it difficult to accurately depict the continuous change process of the scene, and lacking sufficient constraints on computational efficiency and consistency of multi-source information.

Method used

A driving video generation method based on spatiotemporal factorization architecture and hybrid modulation is adopted. By aligning historical multi-view driving image sequences with environmental condition information, latent variable encoding, diffusion modeling, spatiotemporal joint modeling and cross-view feature interaction are performed to generate future multi-view driving video sequences.

Benefits of technology

It improves the spatial consistency, temporal continuity, and multi-view consistency of the generated results, provides more reliable scene information support, and enhances the environmental prediction and behavior planning capabilities of autonomous driving systems.

โœฆ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122093518B_ABST
    Figure CN122093518B_ABST
Patent Text Reader

Abstract

The application is suitable for the field of automatic driving technology, and provides a driving video generation method based on a space-time factorization architecture and hybrid modulation, which comprises the following steps: obtaining a condition set corresponding to each time frame and each view angle, obtaining a diffusion input latent variable, obtaining a space-time fusion feature, and finally obtaining a future multi-view driving video sequence.The application has the advantages that image information, environmental structure information and vehicle control intention are uniformly included in the same generation framework, the expression ability of the model for a complex driving environment is improved, the controllability of the generation process is enhanced while reducing the computational complexity, the stability of the generation result in terms of spatial structure, time continuity and multi-view consistency is improved, and more reliable scene information support is provided for environment prediction, behavior planning, auxiliary decision-making and simulation training and other applications in the automatic driving system.
Need to check novelty before this filing date? Find Prior Art