Video shadow detection method and device based on language driving and related equipment

Through natural language instruction analysis and dual-track collaborative memory technology, combined with self-attention optimization, the flexibility and timing coherence problems in traditional shadow detection technology are solved, and high-accurate video shadow detection is achieved.

CN119206562APending Publication Date: 2024-12-27HONG KONG UNIV OF SCI & TECH (GUANGZHOU)
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202411095240.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-09
Publication Date
2024-12-27

AI Technical Summary

Technical Problem

Traditional shadow detection technology lacks flexibility, timing coherence and semantic understanding, making it difficult to personalize shadow recognition and segmentation according to user needs, and the detection performance in complex scenarios is degraded.

Method used

The language-driven video shadow detection method is adopted to determine the target detection area through natural language instruction analysis, and combined with dual-track collaborative memory and self-attention optimization technology, a hierarchical memory structure is constructed and shadow detection is performed.

Benefits of technology

Intelligent and personalized video shadow analysis is realized, which improves the flexibility and accuracy of shadow detection, can model the performance of shadows on different time scales, and enhances the robustness and coherence of detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119206562A_ABST
    Figure CN119206562A_ABST
Patent Text Reader

Abstract

The invention provides a video shadow detection method and device based on language driving and related equipment, and relates to the field of video processing, and the method comprises the steps: firstly obtaining target video data and a natural language instruction; analyzing the natural language instruction, and determining a target detection area in the target video data; performing shadow detection on each video frame according to the target detection area to obtain an initial shadow detection result corresponding to the target detection area in each video frame; the target video data is divided into a plurality of memory fragments, each memory fragment is divided into a plurality of continuous memory blocks, and each memory block comprises a plurality of continuous video frames; and finally, according to the memory block corresponding to each video frame and the memory segment corresponding to each video frame, optimizing the initial shadow detection result to obtain a target shadow detection result corresponding to the target detection area in each video frame. According to the method, the specific shadow area in the specified video is described through the natural language, and the flexibility of shadow detection can be improved.
Need to check novelty before this filing date? Find Prior Art

Citation Information

Cited By

  • Visual adaptive method and system based on attention mechanism

    CN119815081A