A Fast Method for Predicting Saliency in Panoramic Video

CN120298955BActive Publication Date: 2026-06-30NANJING UNIV OF INFORMATION SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
NANJING UNIV OF INFORMATION SCI & TECH
Filing Date
2025-05-29
Publication Date
2026-06-30

AI Technical Summary

Technical Problem

Existing methods for predicting the saliency of panoramic videos are computationally complex and lack real-time performance, making it difficult to meet the needs of real-time applications. Furthermore, traditional center-prior approaches are ill-suited to the unique spatial distribution and human visual attention characteristics of panoramic videos.

Method used

A lightweight panoramic video saliency prediction architecture is designed. Through a basic visual feature extraction module, a spatiotemporal decoupling module, a dual-branch attention prior module, and a temporal relationship modeling module, an adaptive prior model is constructed to simulate the visual bias phenomenon in panoramic video viewing behavior and generate saliency maps.

Benefits of technology

It significantly reduces computational complexity, improves processing speed and accuracy, meets real-time requirements, captures temporal dependencies and dynamic changes in video sequences, and generates more temporally consistent saliency maps.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN120298955B_ABST
    Figure CN120298955B_ABST
Patent Text Reader

Abstract

This invention provides a fast method for predicting the saliency of panoramic videos. First, the panoramic video is processed into video frames, with consecutive video frames used as input. Addressing the challenges of the large field of view and complex content in panoramic videos, a two-stage cascaded method is designed. First, a basic visual feature extraction module extracts multi-scale semantic features from the input frame sequence to quickly extract useful information and filter background information. Then, a spatiotemporal decoupling module extracts the temporal and spatial features of consecutive frames sequentially. Next, a dual-branch attention prior module generates an equatorial prior and a spatial semantic ensemble prior to simulate the visual bias phenomenon in panoramic video viewing behavior. Finally, the spatiotemporal features and prior features are fused, and a temporal relationship modeling module processes the temporal relationships between consecutive video frames to generate the final saliency map.
Need to check novelty before this filing date? Find Prior Art